• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A method for detecting and tracking targets in a series of successive images includes limiting the number of blocks that are the subject of simultaneous tracking or false tracks. The operation of the tracking module is thus improved without the need to increase a detection threshold pads. The detection threshold can even be reduced, so that the detection range is increased and the tracking of each target is more continuous, without the probability of false alarm itself being increased.
    公开号:FR3017481A1
    申请号:FR1400355
    申请日:2014-02-07
    公开日:2015-08-14
    发明作者:Dominique Maltese；Ahmed Hamrouni
    申请人:Sagem Defense Securite SA；
    IPC主号:
    专利说明:

    [0001] The present invention relates to a method for detecting and tracking ("search and tracking") targets, as well as a system that is adapted to implement such a method. Target detection and tracking is a commonly used mode of environmental monitoring, particularly in the naval sector. It involves successively entering a series of optronic images of the environment, then looking in each image for the potential presence of targets, and confirming or denying such a presence by determining whether a repetition of detections exists for several successive images.
    [0002] Such a monitoring mode is often implemented from images of the environment that are captured in an infrared spectral band. In this case, the detection and tracking system is usually designated by IRST, for "InfraRed Search and Track". In addition, the monitoring of the environment can be panoramic. For this, the environment can be scanned by continually varying the line of sight in azimuth over 360 ° (degree). Each scan passage then produces a new image that is also called a banner. The following terms are used in the technical field of detection and tracking of targets: - plot, or hot spot ("spot" in English): place inside an image where the intensities that are captured for some at less pixels of this place have features that are different from the environment in the image. In particular, the intensities of the pixels of a pad may be greater than those of the near pixels. A pad may correspond to a target that is actually present in the field of view or to a detection aberration. A pad is detected by implementing a detection criterion, which can be set with respect to an average signal level and / or a noise level of the picture. For example, the detection criterion may require that the intensity value at a point in the image, to which the intensity value of the background image in the vicinity of the point can be subtracted, be greater than one. predetermined detection threshold. More generally, the detection may be based on the value of a signal-to-noise ratio at the place in the image where the detection of a pad is tested; - probability of detection: probability that a target that is actually present in the surveillance field is detected in an image as a plot; probability of false alarm: probability that a pad that is detected in an image is a detection aberration, that is to say, so that this pad does not correspond to a target that is actually present in the surveillance field . Such a false alarm pad is tracked in the monitoring method, as is any pad that is detected; - tracking sequence: an algorithm for tracking a target in the form of a spatio-temporal chain of plots detected in several successive images, to result either in a declaration of the chain of plots as corresponding to a real target, a decision that the chain of pads is considered to be a detection aberration; - Probability of launching: probability that a real target will be detected and then declared at the end of the runway; and - false-track probability: the probability that a spatio-temporal chain of detected plots will be wrongly declared as a target at the end of the run. In a known manner, in a target detection and tracking method, the two phases of the method are linked in the following way: the first phase consists of detecting pads in each image, then the tracking phase, or tracking , is performed iteratively for each pad that has been detected. The phase of detection of the pads itself comprises the following two steps: / 1 / using a detection threshold, search for pixels inside the image, each of which has a value of a signal-on-signal ratio noise greater than the detection threshold; and / 2 / within a matrix of the image, grouping some of the pixels whose values of the signal-to-noise ratio are greater than the detection threshold, when said pixels are close to each other in the image, so as to form clustering pads. Optionally, other pixels whose values of the signal-to-noise ratio are also greater than the detection threshold may be deleted if these other pixels are isolated in the image. Such a detection phase is executed for each image individually within an image processing module. It is a spatial filtering of the content of the image, which is applied separately from the other 15 images. The tracking phase consists of searching whether each pad that is detected can be spatially chained to other pads that are present in the successive images entered previously. For this, the tracking sequence is tested for all the pads that are identified in the image that is being processed, or for a reduced selection of some pads. When a block has been detected in the image being processed without being able to correlate with one or more pads that have been detected in images of the series prior to the image being processed, the tracking sequence is initiated for this new plot. But if a pad which is detected in the image being processed can be attached to a track already identified from images of the series prior to the image being processed, the tracking is continued with the sequence of launching already initiated. Such an attachment is made taking into account a possible movement of the target between the image that is being processed and the previous one. A target track is then declared for a spatio-temporal chain of plots if the tracking sequence is completely satisfied for this chain of pads. The tracking algorithm of a pad can be constituted by a sequence of detection reports of the pad in successive images, called the tracking sequence. Such an algorithm is assumed to be known to those skilled in the art, and described for example in WO 02/21152. A detection ratio is M / N, where M and N are two positive and non-zero integers, and M is less than or equal to N. The detection ratio means that the chain of pads is present in M images among N successive images . The detection report is satisfied if the chain of the pads is detected in at least M of the N successive images. The tracking sequence is itself satisfied if all the detection reports of this sequence are satisfied. In this case, the setting to the track results in the declaration of the chain of pads as being a target. As soon as one of the detection reports is not satisfied, the chain of pads is considered as a detection aberration and is discarded. In practice, the tracking algorithm is effective if the number of blocks that are tracked in parallel in successive images is not too high. Indeed, a large number of blocks that are tracked simultaneously may require a computation time incompatible with the duration between two images that are seized successively, or require a quantity of memory and computing power that are not available in the system used for monitoring. Moreover, a large number of blocks may reveal trajectories of pads that are close to each other, or that intersect, so that indeterminacy appears as to the allocation of portions of the trajectories to a chain of pads or a another for the tracking of two different chains of studs. For these reasons, the tracking of the detected blocks is more effective when the number of these pads remains limited, that is to say when the probability of false alarm is reduced. Usually, such a limitation on the number of pads that are detected is achieved by adjusting, in the sense of an increase, the threshold that is used in the detection step. The probability of false alarm is reduced, but such an increase in the detection threshold has the effect of reducing the sensitivity of the entire process. In particular the maximum distance of a potential target up to which the detection is really effective, is reduced. This maximum distance is called detection range ("detection haul" in English) in the jargon of the skilled person. Another disadvantage of increasing the detection threshold is the risk of not detecting in a newly processed image a target that has already been detected one or more times in previous images. Accumulation of such sensing holes may result in an interruption of the track chain tracking sequence, thereby reducing the probability of being tracked. Several methods are already known to reduce the probability of false alarm without reducing the scope of the detection of the pads.
    [0003] A first of these methods is called "track before detect" in English, or "continuation before detection". It consists in looking for a confirmation in several successive images, at least some of the points whose intensity or signal-to-noise ratio is greater than the detection threshold, before qualifying them or not as pads and initiating a tracking for each of them. This method is very expensive in memory and calculation time because it intervenes in the image processing, comparing individual intensities of image points between at least two successive images. A second known method is to filter the pads that are detected according to individual attributes of each pad within the image in which it is detected for the first time. For example, the number of adjacent image points that together make up the plot, the shape of the plot outline, or morphology of the pad, and its radiometric intensity are attributes that can be taken into account in filtering. Tracking is then performed only for the pads that have been retained after such filtering. The article "On Track-Management within the PMHT Framework" by Monika Wieneke and Peter Willett, ISIF FUSION Symposium 2008, describes a third method to reduce the probability of false alarms, as part of a detection and tracking process which is "measure-oriented". In such a method, a probabilistic multiple-hypothesis tracking module, or PMHT ("Probabilistic Multiple - 6 - Hypothesis Tracker"), can process in parallel several trajectories that are possible for a potential target, during its tracking, from the positions of blocks that are identified in successive images. In such a method of detection and tracking, the pads that are detected in a newly captured image are first affected as much as possible to the trajectories already initiated from previous images. Then, the residual pads, that is to say those which could not be thus assigned to a hypothetical trajectory already identified, are filtered before initiating a new tracking sequence for each of them. Such filtering is performed by an additional module of the target detection and tracking system, which is called "GMPHD Clutter Remover", or hypothetical probability density clutter module based on Gaussian mixture ("Gaussian-Mixture Probability Hypothesis Density ". Such a module constructs a probability density distribution within an image matrix, from a current state of tracking sequences and trajectories of the pixels that result from images prior to the image. which is being processed. Such a distribution of probability density, which is also called "intensity function" by the skilled person, is constructed by applying kinematic criteria of displacement of the pads between different images, and updated for each new image that is captured. then treated. Thus, target detection and tracking methods are also known, which further include the steps of: / 3 / selecting some of the pads that have been formed in step / 2 /, in restricted numbers relative to the set of pads formed at this step / 2 /, and / 4 / test a tracking sequence only for the selected pads, and declare a target track for a chain of selected pads in different frames if the setting sequence track is completely satisfied for this chain of studs.
    [0004] Compared to these known methods, a first object of the invention is to increase the efficiency of the tracking with respect to the image capture frequency, without reducing the detection range. A second object of the invention is to reduce the probability of false alarm without reducing the probability of placing on the track. An additional object of the invention is to achieve these first and second goals without requiring additional amount of memory and computing power that are too important. Finally, another complementary object of the invention lies in the use of a detection and tracking method which is "track-oriented" ("track-oriented" in English). To achieve these or other objects, a first aspect of the present invention provides a method for detecting and tracking targets in a series of optronic images of the same monitoring field, which includes steps / 1 / to / 4 / performed for each image, but in which the number of pads which are selected in step / 3 / is limited according to at least one statistical characteristic relating to the tracking sequences which are already being tested before that this step / 3 / is executed for the image being processed. Thus, the effective number of the traces which are continued in parallel at the same time of the process is limited, so that a saturation of the tracking capacity is avoided. In addition, this limitation reduces the risk that trajectories of different track chains that are tracked overlap or intersect. A situation of indeterminacy, also called ambiguity by the skilled person, in the assignment of the portions of trajectories to the chains of studs concerned is therefore avoided more often. In addition, the probability of false alarm is reduced without it being necessary to increase the threshold of detection of the pads inside each image. The range of detection within the surveillance field is therefore not reduced. Similarly, not increasing the detection threshold makes it possible to reduce the risk that a block corresponding to a chain already engaged does not appear in a subsequent image, by causing a detection hole, or even a tracking interruption, while this Current string of studs corresponds to a real target that has been continuously in the monitoring field. The probability of placing on the track is therefore maintained or improved by the invention. Such a method according to the invention can be "track-oriented" because a pad that is newly detected does not call into question a track already in progress with the potential target trajectory associated with it. In preferred embodiments of the invention, the number of pads which are selected in step / 3 / may be limited according to an integrated intensity value Nc which is calculated for a density distribution of target presence probability, produced for the tracking sequences that are already being tested before step / 3 / is executed for the image being processed, and based on a number of Nt pads for which the tracking sequences are already being tested, and which have individual weights greater than a predetermined weight threshold. In other words, the statistical characteristic relating to the tracking sequences that are already being tested before the step / 3 / is executed for the image being processed, and which is used to select the pads during of the execution of step / 3 / for the image being processed, includes the integrated intensity value Nc and the number of pads Nt. These two components of the statistical characteristic are readily available without requiring additional calculations that are important. For some of these preferred modes of implementation of the invention, the weight threshold which is used for the individual weight of each pad whose tracking sequence is being tested, in order to determine the number of Nt pads, can be equal to 0.5.
    [0005] Preferably, the number of pads which are selected in step / 3 / may be between the number of pads Nt and the integrated intensity value Nc increased by a predetermined margin Mf, or be equal to the number of pads Nt or the integrated intensity value Nc plus the margin Mf. Even more preferably, the number of pads which are selected in step / 3 / may be equal to the smaller of the number of pads Nt and the integrated intensity value Nc plus the margin Mf. In those of the preferred embodiments of the invention which use the margin Mf, this may be a non-zero integer which is typically less than or equal to five. The method of the invention may further comprise the following steps: - producing the target presence probability density distribution from a current state of the tracking sequences that have been initiated for images prior to image being processed, and from the trajectories of the pads which correspond to these initiation sequences initiated for previous images, by applying kinematic criteria of displacement of the pads between successive images; - find a correlation between the target presence probability density distribution and the pads that were formed in step / 2 /; then - update the target presence probability density distribution using a result of the correlation. The statistical characteristic relating to the tracking sequences which are already being tested when the step / 3 / is executed for the image being processed, and which is used to limit the number of the pads selected at this step / 3 / can advantageously be obtained from the updated target presence probability density distribution. Thus, a finishing sequence that is completed can be directly taken into account to initiate a new tracking sequence from another selected plot in the image being processed.
    [0006] In particular implementations of the invention, each pad may have a Gaussian profile within the target presence probability density distribution. Finally, in general, the pads that are selected in step / 3 / may correspond to tracking sequences that are already under test, or may have weight values that are greater than the weight values of non-selected pads, or may have respective signal-to-noise ratio values that are greater than a selection threshold, or may each include a number of image pixels that is greater than a minimum pad size. , or may satisfy criteria that are related to a type assumed for the targets.
    [0007] A second aspect of the invention provides a target detection and tracking system which comprises the following components: - a device for capturing optronic images; an image processing module, which is adapted to calculate respective values of a signal-to-noise ratio for pixels of each captured image, to group together some of the pixels whose signal-to-noise ratio values are greater than a detection threshold, when these pixels are close to each other in the image captured, so as to form clustering pads, and possibly also to remove others pixels whose signal-on-signal values noise are also greater than the detection threshold, if these other pixels are isolated in the image; a tracking module, which is adapted to test a tracking sequence for at least some of the pads, by using a chain of pins detected in different images, and to declare a target track for the chain of pads if the sequence the track is satisfied for this chain of studs; and a module for selecting pads, which is arranged to receive, for each image captured, the pads formed by the image processing module, and to output some of the pads received at the input, and arranged so that the tracking module only tests the tracking sequence only for those pads that are output by the selection module. The system of the invention is characterized in that the selection module is adapted to select the pads which are outputted by limiting a number of these pads outputted as a function of at least one statistical characteristic, which relates to Tracking sequences that are already being tested before the pads are selected for the image being processed. Such a system is adapted to implement a method according to the first aspect of the invention, including in its preferred embodiments that have been cited. In particular, the system may further include a hypothetical probability density filter, which is adapted to produce the target presence probability density distribution from the current state of the tracking sequences established by the module. tracking, and from the trajectories of the pads that correspond to these tracking sequences, by applying kinematic criteria of displacement of the pads between successive images, and to search for the correlation between the target presence probability density distribution and the pads that have been formed by the image processing module, and then to update the target presence probability density distribution using a result of the correlation. In this case, the hypothetical probability density filter may incorporate the pad selection module. In particular, the hypothetical probability density filter can be adapted to calculate itself from the updated target presence probability density distribution, the integrated intensity value Nc and the number of Nt pads for which the tracking sequences are being tested, and have individual weights greater than the weight threshold. In preferred embodiments of the invention, the hypothetical probability density filter may be of the combination type of Gaussian profiles.
    [0008] Other features and advantages of the present invention will appear in the following description of nonlimiting exemplary embodiments, with reference to the accompanying drawings, in which: - Figure 1 shows the main components of a detection system and target tracking which is in accordance with the invention; and FIG. 2 symbolically illustrates a target presence probability density distribution, as may be used in a method according to the invention. By way of illustration, the invention is now described for an operational monitoring system in the infrared spectral range. In this case, and with reference to FIG. 1, the system may be of the IRST type and include the image capture device 10 with the processing and analysis modules 11 to 13. It may further comprise additional components, such as image storage memories and controllers of such memories, but the use of such additional components in a target detection and tracking system is assumed to be known, and is not repeated here. The device 10 can be mounted on a rotating turret about a vertical axis, and controlled to continuously perform rotation turns at a constant speed. In this way, an image can be captured at each rotation turn, which represents the limited system environment between a minimum value and a maximum height value in elevation. Each shot during a complete turn constitutes a new image, so that a series of successive images is captured during a continuous rotation of the turret.
    [0009] Thus, the acquisition frequency of the images can be between 0.1 and 1 Hz (hertz), for example. Other alternative modes of capturing images, for example using a device 10 that is stationary, can be used alternately. The capture rate of the images can then be higher, especially greater than 10 Hz.
    [0010] Each image that is captured by the device 10 is transmitted to an image processing module 11. This module 11 has the function of extracting pads of each image, independently of other images that have been previously captured, and to transmit the coordinates and possibly a value of the signal-to-noise ratio of each pad to a hypothetical probability density filter 12. Currently, the image processing module 11 comprises at least two distinct units, which are respectively referenced 1 and 2 The unit 1 is dedicated to the calculation of a value of a signal-to-noise ratio, denoted S / B, individually for at least some of the pixels constituting a new image captured by the device 10. Preferably, the S / N ratio is calculated for each pixel of an image captured. It can be calculated in one of the ways known to those skilled in the art, for example by using an overall evaluation of the noise on the entire image, or by using a local evaluation of the image noise around each pixel, and possibly by subtracting a background level of image intensity. Several evaluation modes and improvements known before the present invention can be implemented to calculate the S / N ratio.
    [0011] The unit 1 also receives a detection threshold, which is compared the value of the S / N ratio of each pixel. The detection threshold can be absolute or defined in relative value with respect to a maximum value that is reached in the image for the S / N ratio. The unit 1 then transmits to the unit 2 the coordinates of the pixels for which the values of the S / N ratio are greater than the detection threshold, and possibly the values of the S / N ratio themselves. Unit 2 groups the pixels transmitted by unit 1 according to their proximity to the inside of the image. Pixels for which the value of the S / N ratio is greater than the detection threshold, and which are adjacent or close to each other in the image, are gathered in a grouping pad, simply designated by pad in the following sequence. this description. Each block is identified and associated with its coordinates in the image, and possibly also with a global value of the S / N ratio which is defined for this plot. Plot formation algorithms, which can be implemented in unit 2, are also well known to those skilled in the art, so that it is not necessary to describe them again here. Optionally, such algorithms can simultaneously eliminate pixels for which the individual values of the S / N ratio are likewise greater than the detection threshold, but which are isolated in the image. Such isolated pixels may also include small groups of neighboring pixels with individual values of the S / N ratio which are greater than the detection threshold, but which would form clustering blocks whose overall S / N values for each of these small ones. groups would be too weak. The image processing module 11 then transmits to the module 12 the identification and the coordinates of the pads, and possibly the values of the S / B ratio determined for each pad. Each pad that is outputted by the image processing module 11 can also be associated with a weight that results from a combination of a spatial extension of the pad in the image with a contrast of this pad. In other words, the module 11 performs a spatial filtering of the content of each image that is captured by the device 10, and the areas of the image where the filtered intensity is greater form the pads that are delivered to the module 12. The module 12 may be of the "PHD filter" type for "Probabilistic Hypothesis Density", and the module 13 is dedicated to the tracking of targets within several images that have been seized successively. The principle of tracking may be that which is already known before the present invention, and which has been described above. The modules 12 and 13 operate together to assign some of the pads that are delivered by the module 11 for a new image, in pursuit of trajectories in the array of images that result from pads detected in previous images. These trajectories are therefore formed of spatio-temporal chains of pads that have been detected in different images. To extend a chain already initiated, the module 12 itself comprises a prediction unit 3 and an update unit 4. The prediction unit 3 constructs a target presence probability density distribution from the sequences of which have been initiated or confirmed for earlier images and which are still being tested or maintained, and coordinates in the image matrix of the pads concerned by these sequences. This probability density distribution is commonly called the intensity function, and assigns to each point of the image field, or to each pixel of the image matrix, a probability density value for a target to actually be present at that location. . For this, pads that have been detected in several images are assigned to the same trajectory in accordance with a continuity constraint of the trajectory between successive images. The trajectory is then extrapolated for the moment of capture of the image that is being processed, that is to say the last image for which the module 11 provided the detected pads. This extrapolation can also take into account kinematic constraints related to a type of target that is assumed, such as a maximum displacement speed, for example. In addition, each block in the probability density distribution has an individual weight whose value is higher as the tracking sequence is already followed for a significant number of units. images, or that the values of the S / N ratio of the pads of this mise en piste in the successive images are high, or that these values of the S / N ratio have a regular evolution, etc. Different methods, or combinations of several methods, for assigning a weight to each plot in the probability density distribution are also assumed to be known. The module 4 is dedicated to updating the probability density distribution as provided by the prediction unit 3. For this, those of the blocks transmitted by the image processing module 11, which are substantially superimposed probes of the probability density distribution, or close to such studs, are merged with them. Thus, the probability density distribution is modified, i.e. updated, to account for the information that is extracted from the image being processed. This update concerns the position of each pad in the image matrix, its spreading and its weight. Algorithms that are adapted to achieve such an update are also known to those skilled in the art and are not part of the contribution of the present invention. Figure 2 symbolically represents a target presence probability density distribution, denoted P, as may be used in the invention. The distribution P can have values for all points of the matrix common to all the images that are captured by the device 10. It can be defined by a sum of envelopes each with an amplitude, one or two transverse dimensions and coordinates of a center of this envelope inside the image matrix. The envelopes may each have a predetermined shape, for example a Gaussian shape. Each envelope is adjusted for its parameters to coincide with a pad, and the amplitude of the envelope is determined by the weight of the pad. The references 100 denote pads whose weights are important in the distribution of probability density P, and references 200 denote pads of lower weight. The coordinates of each plot in the image array are the column number noted C and the line number noted L, which correspond to the center of the plot. In the case of Gaussian envelopes, the filter 12 is said to be of the "GMPHD" type for "gaussian-mixture probabilistic hypothesis density". To implement the present invention, statistical characteristics can be calculated from the probability density distribution, possibly by the update unit 4 itself. For example, such features may include an integrated intensity value for the probability density distribution. Integral intensity value, denoted Nc, is an integral sum of the individual values of the probability density distribution for all the points of the image matrix. Nc is a number of different targets that are suspected in the monitoring field, based on the current state of the tracking sequences. Another statistical characteristic of the probability density distribution may be the number of pads in this distribution that have a weight greater than a predetermined weight threshold. Such a weight threshold can be defined as a fixed value, or a relative value, such as for example a fraction of a maximum plot weight which is reached in the probability density distribution, or a fraction of an average weight. plots that are present in the distribution. For example, the weight threshold can be set at 0.5, or 0.4, or else 0.25. The object of the invention is to limit the number of launches which are continued simultaneously and in parallel by the tracking module 13. In other words, operation is sought for the module 13, which is a compromise between the processing capacity of this module and the complexity of the probability density distribution for the series of images that is analyzed. The processing capacity of the module 13 is the number of launches that can be tracked in parallel for a determined period of analysis of each image. For this, the filter 12 makes a selection among the pads that are transmitted by the image processing module 11 for the image that is being processed. This selection can be performed by a dedicated module, referenced 5. The selection of the pads is limited by the number of pads that will actually be transmitted to the tracking module 13. The pads not transmitted -17- can be abandoned permanently, or stored temporarily. In a preferred mode of implementation of the invention, but which is not limiting, this number of pads that are transmitted to the tracking module 13 may be equal to the smaller of the following two values: the intensity value integrated Nc of the probability density function, increased by a margin Mf: Nc + Mf, and the number Nt of the pads that have individual weights greater than the predetermined weight threshold, in the updated probability density function.
    [0012] For example, the margin Mf may be a non-zero integer less than or equal to ten, preferably less than or equal to five, for example equal to two or three. The value that is chosen for the margin Mf controls the average number of new launcher sequences that can be initiated at each new image that is analyzed, or the average frequency at which such new launcher sequences are initiated, by compared to the number of images analyzed. Thus, for a distribution of probability density that is not very complex, which comprises few pads but these being each important, the number of the pads of the image currently being processed which are selected to be transmitted to the tracking module 13 is Nt . These selected pads of the image being processed may include some of those that coincide with the pads of the updated probability density function, but not necessarily all of them. New pads may thus be selected, depending on the number of detection holes that are introduced into the tracking sequences under test, or the number of such sequences that eventually terminate. A new tracking sequence is then initiated for each of these new pads. This case of distribution of probability density not complex is the most common in practice. When the probability density distribution becomes more complex, it comprises significant pads, that is to say whose individual weights are each greater than the predetermined weight threshold, in greater numbers. The number of pads which are selected by the module 5 to be transmitted to the tracking module 13 then increases until it is limited to Nc + Mf. Overload is thus avoided for operation of the tracking module 13. In this case again, some of the pads that are selected may not correspond to any of the tracking sequences that are already being tested. The pads that are transmitted by the module 5 to the module 13, in number which is limited in the manner just described, can be selected in several ways, according to criteria that are used for this selection. Priority is not necessarily given to pads for which tracking sequences are already being tested. Possible criteria for selecting the pads that are transmitted to the tracking unit 13 may be: the pads are selected in descending order of their respective weight values, or those pads whose S / N value is greater than a predetermined selection threshold, or those of the pads that satisfy one or more criteria related to a type assumed for the targets. Using a combination of several of these criteria within the module 5 to select the pads that are transmitted to the tracking module 13 is possible and may be advantageous. Criteria that are related to the type of targets may relate to target morphology, apparent movement rates, and the like. The selection of the pads which is thus carried out by the module 5 produces a reduction of the clutter present in the processed image. The tracking module 13 then continues the tracking sequences that are already in progress, and possibly initiates new tracking sequences, only for those of the pads that have been selected. The testing and updating of each sequence by the module 13 are known per se and are not modified by the present invention. The declared tracks can then be submitted to a monitoring operator, who validates or not the detection of a real target. The following table compares monitoring performance obtained with a target detection and tracking system that is devoid of the selection module 5, and another system that incorporates such a selection module used in accordance with the invention. Two values for the detection threshold, respectively indicated by the mean threshold and the low threshold, are furthermore used for the system that does not have a selection module, whereas the system according to the invention is set implemented with the low threshold. The same scene of the appearance of an aircraft in the surveillance field, from an edge of the surveillance field, then from the airplane in a cloudy sky while remaining inside the monitoring, is used for the three experiments. Without modulus of Without module of With modulus of selection - Average threshold selection - Low threshold selection Low threshold Detections Images 794 to 1285 then 1385 to 1415 Images 794 to 1068 then 1310 to 1461 Images 794 to 1461 positive of the plane Number of Average = 1 , 81 SD = 2.07 Mean = 0.497 false runs per image SD = 0.696 Number of false runs created per image Mean = 0.046 SD = 0.225 Mean = 0.013 SD = 0.117 Number of false runs for clouds per image Mean = 0.212 Standard deviation = 0.409 Mean = 0.995 Standard deviation = 0.136 Number of false runs for created clouds per image Average = 0.009 Standard deviation = 0.096 Mean = 0.005 Standard deviation = 0.073 The number of false leads is therefore greatly reduced by the use of the selection module 5 according to the invention. In addition, the fact that the continuation of the aircraft continues to image 1461 for the last two columns of the table, instead of stopping at image 1415, shows that the tracking of the aircraft is preserved and continuous. In addition, the positive detections of the aircraft show that the reduction of the detection threshold did not introduce false detections. In addition, for cloud edges, the clouds are detected and tracked at 99.5% of the monitoring time when the selection module 5 is used according to the invention, whereas they are detected and tracked only when at 21.2% of the monitoring time for the average detection threshold without using the selection module 5. The increase in the number of images in which the cloud edges are detected and tracked, provides the operator with a time which is sufficient to more easily identify the targets concerned as the edges of the clouds. In addition, being stable over time, once identified, these targets are no longer disruptive for the operator.
    [0013] Thanks to the use according to the invention of the selection module 5, it has therefore been possible to lower the detection threshold, that is to say to increase the detection sensitivity of a potential target to the inside each captured image, without reducing the probability of being tracked. More specifically, lowering the detection threshold makes it possible to continue for a longer time a target such as the airplane in the preceding example. It also makes it possible to delete certain false tracks while having improved continuity for the detection of a real target in the successive images. It is understood that the invention may be reproduced by introducing modifications with respect to the implementation which has been described in detail above. In particular, the statistical characteristic of the tracking sequences that are already being tested when processing a new image, and which is used to limit the number of pads that are selected in this new image, may be different from the integrated intensity value Nc and the number of significant pads Nt. Moreover, even when this value Nc and this number Nt are used, another combination than the choice of the smallest between the number Nt and the sum of Nc with the margin Mf can be applied to limit the number of the pads which are selected. to be transmitted to the tracking module.
    权利要求:
    Claims (14)
    [0001]
    REVENDICATIONS1. A method for detecting and tracking targets in a series of optronic images of the same monitoring field, comprising the following steps performed for each image being processed: / 1 / using a detection threshold, searching for pixels at inside the image, which each have a value of a signal-to-noise ratio higher than the detection threshold; / 2 / within a matrix of the image, grouping some of the pixels whose signal-to-noise ratio values are greater than the detection threshold, when said pixels are close to each other in the image , so as to form clustering pads; / 3 / select some of the pads formed in step / 2 /, in a restricted number with respect to all the pads formed in said step / 2 /, and / 4 / test a tracking sequence only for the pads selected, and declare a target track for a chain of selected pads in different frames if the tracking sequence is completely satisfied for said chain of pads; the method being characterized in that the number of the pads selected in step / 3 / is limited as a function of at least one statistical characteristic relating to the tracking sequences which are already being tested before said step / 3 / is executed for the image being processed.
    [0002]
    The method of claim 1, wherein said at least one statistical characteristic used to limit the number of pads that are selected in step / 3 / comprises an integrated intensity value Nc calculated for a probability density distribution of target presence, produced for the tracking sequences that are already being tested before step / 3 / is executed for the image being processed, and based on a number of Nt pads for which the tracking sequences are already under test, and have individual weights above a predetermined weight threshold.
    [0003]
    3. Method according to claim 2, wherein the weight threshold which is used for the individual weight of each pad whose tracking sequence is being tested, in order to determine the number of pads Nt, is equal to 0, 5.
    [0004]
    4. The method of claim 2 or 3, wherein the number of pads that are selected in step / 3 / is between the number of pads Nt and the integrated intensity value Nc increased by a predetermined margin Mf, or equal to said number of pads Nt or to said integrated intensity value Nc increased by the margin Mf.
    [0005]
    5. The method of claim 4, wherein the number of pads that are selected in step / 3 / is equal to the smaller between the number of pads Nt and the integrated intensity value Nc increased by the margin Mf.
    [0006]
    6. The method of claim 4 or 5, wherein the margin Mf is a non-zero integer less than or equal to five.
    [0007]
    The method of any one of claims 2 to 6, further comprising the steps of: producing the target presence probability density distribution from a current state of the initiated tracking sequences for images prior to the image being processed, and from trajectories of the pads corresponding to said initiation sequences initiated for prior images, by applying kinematic criteria of displacement of the pads between successive images; - find a correlation between the target presence probability density distribution and the pads that were formed in step / 2 /; then - updating the target presence probability density distribution using a result of the correlation, and according to which the statistical characteristic relating to the tracking sequences which are already being tested when the step / 3 / is executed for the image being processed, and which is used to limit the number of the selected pads in said step / 3 /, is obtained from the updated target presence probability density distribution .
    [0008]
    The method of any one of claims 2 to 7, wherein each pad has a Gaussian profile within the target presence probability density distribution.
    [0009]
    9. A method according to any one of the preceding claims, wherein the pads which are selected in step / 3 / correspond to tracking sequences already under test, or have weight values greater than values. of non-selected pads, or have respective values of the signal-to-noise ratio greater than a selection threshold, or each comprise a number of picture pixels which is greater than a minimum pad size, or satisfy criteria related to a type assumed for the targets.
    [0010]
    10. Target detection and tracking system comprising the following components: - an optronic image capture device (10); an image processing module (11), adapted to calculate respective values of a signal-to-noise ratio for pixels of each captured image, for grouping some of the pixels whose signal-to-noise ratio values are greater than a detection threshold, when said pixels are close to each other in the captured image, so as to form clustering pads; a tracking module (13), adapted to test a tracking sequence for at least some of the pads using a chain of pads detected in different frames, and to declare a target track for the chain of pads if the sequence of setting on track is satisfied for this chain of studs; and 301 7 4 8 1 - 24 - - a module of selection of pads (5), arranged to receive as input, for each image captured, the pads formed by the image processing module (11), and to deliver in outputting some of the pads received at the input, and arranged so that the tracking module (13) tests the tracking sequence only for those pads that are output by said selection module (5), the characterized in that the selection module (5) is adapted to select the pads that are output by limiting a number of said pads outputted as a function of at least one statistical characteristic relating to the setting sequences on the track which are already being tested before the pads are selected for the image being processed.
    [0011]
    11. System according to claim 10, adapted to implement a method according to any one of claims 2 to 6.
    [0012]
    The system of claim 11, further comprising: - a hypothetical probability density filter (12), adapted to produce the target presence probability density distribution from a current state of the implementation sequences; track established by the tracking module (13), and from trajectories of pads 20 corresponding to said tracking sequences, by applying kinematic criteria of displacement of the pads between successive images, and to seek a correlation between the distribution of target presence probability density and the pads that have been formed by the image processing module (11), then to update the target presence probability density distribution using a result of the correlation ; and the hypothetical probability density filter (12) incorporating the pad selection module (5).
    [0013]
    The system of claim 12, wherein the hypothetical probability density filter (12) is adapted to calculate from the updated target presence probability density distribution, the intensity value integrated Nc and the number of Nt pads for which the tracking sequences are being tested, and which have individual weights greater than the weight threshold.
    [0014]
    The system of claim 12 or 13, wherein the hypothetical probability density filter (12) is of a Gaussian profile combination type.
    类似技术:
    公开号 | 公开日 | 专利标题
    EP3103102B1|2018-08-29|Method for detecting and tracking targets
    CA2100107C|2004-01-06|Process and device for measuring precipitation on ground area
    EP2368135B1|2015-11-25|Method for filtering the radar echoes produced by wind turbines
    EP0255177A1|1988-02-03|Automatic extraction process of a contrasted object in a digital image
    EP0745866A1|1996-12-04|Method and apparatus for movement detection of a target and its applications
    WO2014135404A1|2014-09-12|Method for tracking a target in an image sequence, taking the dynamics of the target into consideration
    FR2717269A1|1995-09-15|Discriminating between targets which might represent ocean mines
    FR3050538A1|2017-10-27|METHOD FOR OPTIMIZING THE DETECTION OF MARINE AND RADAR TARGETS USING SUCH A METHOD
    EP2294441B1|2012-02-22|Method of detecting an object in a scene comprising artefacts
    EP0821322A1|1998-01-28|Method for temporal filtering of the noise in an image from a sequence of digital images and apparatus therefor
    FR2706211A1|1994-12-16|Method of processing road images and of monitoring obstacles, and device for implementing it
    FR3048316B1|2019-06-28|DEVICE FOR DETECTING A LASER SPOT
    EP3324361B1|2020-08-05|Method for detecting and tracking targets
    FR2548347A1|1985-01-04|PROCESS FOR PRODUCING AN IGNITION SIGNAL AND DETECTION AND IGNITION DEVICE WITH SEVERAL DETECTORS
    CA2965323A1|2016-05-12|Method for detecting defective pixels
    FR3074342B1|2019-11-08|METHOD FOR DETECTING AND TRACKING TARGETS
    FR2975807A1|2012-11-30|DETECTION AND TRACKING OF TARGETS IN A SERIES OF IMAGES
    EP1306803A1|2003-05-02|Medical imaging apparatus using fast image segmentation
    EP3088272B1|2020-11-25|Method for detecting the position of guardrails in a chairlift through image analysis
    FR3108733A1|2021-10-01|A method of generating a stream of sea clutter, programmable device and associated computer program.
    EP3088271B1|2020-11-25|Method for detecting the position of guardrails in a chairlift through image analysis
    EP3213112A1|2017-09-06|Method for processing a radar signal in land/sea detection mode; processing system and associated computer program product
    EP0740165B1|2000-08-30|Method and device for signal processing for removing ambiguity in a Doppler radar
    FR3072183A1|2019-04-12|METHOD FOR DETECTING RADAR SIGNALS
    FR3097973A1|2021-01-01|RADAR DEVICE FOR THE DETECTION OF REFERENCE BEHAVIOR OF TRACKED TARGETS; PROCESS AND PRODUCT ASSOCIATED COMPUTER PROGRAM
    同族专利:
    公开号 | 公开日
    AU2015215162A1|2016-08-18|
    WO2015117833A1|2015-08-13|
    US10078903B2|2018-09-18|
    AU2015215162B2|2019-11-21|
    IL247091A|2019-11-28|
    IL247091D0|2016-09-29|
    EP3103102B1|2018-08-29|
    EP3103102A1|2016-12-14|
    US20160350938A1|2016-12-01|
    FR3017481B1|2016-02-26|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    WO2011114086A1|2010-03-17|2011-09-22|Isis Innovation Limited|A method of tracking targets in video data|
    FR2389865B1|1977-05-06|1981-11-20|Realisa Electroniques Et|
    US5506682A|1982-02-16|1996-04-09|Sensor Adaptive Machines Inc.|Robot vision using targets|
    US4497065A|1982-07-12|1985-01-29|Westinghouse Electric Corp.|Target recognition system enhanced by active signature measurements|
    US5923776A|1996-05-23|1999-07-13|The United States Of America As Represented By The Secretary Of The Navy|Object extraction in images|
    US6201236B1|1997-11-13|2001-03-13|Auto Sense Ltd.|Detection system with improved noise tolerance|
    US6369749B1|2000-09-07|2002-04-09|Lockheed Martin Corporation|Adaptive control of the detection threshold of a binary integrator|
    US8958654B1|2001-04-25|2015-02-17|Lockheed Martin Corporation|Method and apparatus for enhancing three-dimensional imagery data|
    IL155034D0|2003-03-23|2004-06-20|M A M D Digital Data Proc Syst|Automatic aerial digital photography and digital data processing systems|
    US20060245500A1|2004-12-15|2006-11-02|David Yonovitz|Tunable wavelet target extraction preprocessor system|
    US7414706B2|2004-12-22|2008-08-19|Northrop Grumman Corporation|Method and apparatus for imaging a target using cloud obscuration prediction and detection|
    US9778351B1|2007-10-04|2017-10-03|Hrl Laboratories, Llc|System for surveillance by integrating radar with a panoramic staring sensor|
    US8526717B2|2007-05-15|2013-09-03|Vision Interface Technologies, LLC|Rich color transition curve tracking method|
    JP5097480B2|2007-08-29|2012-12-12|株式会社トプコン|Image measuring device|
    EP2312994B1|2008-07-18|2021-01-27|Doheny Eye Institute|Optical coherence tomography - based ophthalmic testing systems|
    DE102009009896B4|2009-02-20|2011-02-10|Eads Deutschland Gmbh|Method and device for detecting target objects|
    JP5469899B2|2009-03-31|2014-04-16|株式会社トプコン|Automatic tracking method and surveying device|
    CA2899584A1|2013-01-29|2014-10-23|Andrew Robert Korb|Methods for analyzing and compressing multiple images|
    FR3017481B1|2014-02-07|2016-02-26|Sagem Defense Securite|METHOD FOR DETECTING AND TRACKING TARGETS|JP5919665B2|2011-07-19|2016-05-18|日本電気株式会社|Information processing apparatus, object tracking method, and information processing program|
    FR3017481B1|2014-02-07|2016-02-26|Sagem Defense Securite|METHOD FOR DETECTING AND TRACKING TARGETS|
    US10026015B2|2014-04-01|2018-07-17|Case Western Reserve University|Imaging control to facilitate tracking objects and/or perform real-time intervention|
    US11175142B2|2014-07-31|2021-11-16|Honeywell International Inc.|Updating intensities in a PHD filter based on a sensor track ID|
    US10309784B2|2014-07-31|2019-06-04|Honeywell International Inc.|Merging intensities in a PHD filter based on a sensor track ID|
    US10605607B2|2014-07-31|2020-03-31|Honeywell International Inc.|Two step pruning in a PHD filter|
    US10248839B2|2015-11-30|2019-04-02|Intel Corporation|Locating objects within depth images|
    US10810442B2|2016-09-13|2020-10-20|Nec Corporation|People flow estimation device, people flow estimation method, and recording medium|
    FR3074342B1|2017-11-29|2019-11-08|Safran Electronics & Defense|METHOD FOR DETECTING AND TRACKING TARGETS|
    CN109886305B|2019-01-23|2021-05-04|浙江大学|Multi-sensor non-sequential measurement asynchronous fusion method based on GM-PHD filtering|
    CN111736145A|2020-06-28|2020-10-02|电子科技大学|Multi-maneuvering-target Doppler radar tracking method based on Gaussian mixed probability hypothesis density filtering|
    法律状态:
    2015-02-20| PLFP| Fee payment|Year of fee payment: 2 |
    2016-01-21| PLFP| Fee payment|Year of fee payment: 3 |
    2017-01-24| PLFP| Fee payment|Year of fee payment: 4 |
    2017-03-03| CD| Change of name or company name|Owner name: SAGEM DEFENSE SECURITE, FR Effective date: 20170126 |
    2018-01-23| PLFP| Fee payment|Year of fee payment: 5 |
    2019-01-23| PLFP| Fee payment|Year of fee payment: 6 |
    2020-01-22| PLFP| Fee payment|Year of fee payment: 7 |
    2021-11-12| ST| Notification of lapse|Effective date: 20211005 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1400355A|FR3017481B1|2014-02-07|2014-02-07|METHOD FOR DETECTING AND TRACKING TARGETS|FR1400355A| FR3017481B1|2014-02-07|2014-02-07|METHOD FOR DETECTING AND TRACKING TARGETS|
    US15/117,071| US10078903B2|2014-02-07|2015-01-22|Method for detecting and tracking targets|
    PCT/EP2015/051244| WO2015117833A1|2014-02-07|2015-01-22|Method for detecting and tracking targets|
    EP15700756.8A| EP3103102B1|2014-02-07|2015-01-22|Method for detecting and tracking targets|
    AU2015215162A| AU2015215162B2|2014-02-07|2015-01-22|Method for detecting and tracking targets|
    IL24709116A| IL247091A|2014-02-07|2016-08-03|Method for detecting and tracking targets|
    [返回顶部]