Target detection device and target detection method

Abstract

This target detection device is provided with: an amplitude likelihood calculation unit (101) that, on the basis of a plurality of consecutive frames, calculates an amplitude likelihood for each frame; a tracking processing unit (105) that calculates a tracking filter estimate by performing a tracking process for a target using N tracking filters, where N is a number smaller than the number of resolution cells in each frame; a transition probability calculation unit (102) that calculates a transition probability of the target from the previous frame to the current frame on the basis of the frames and the result of the calculation by the tracking processing unit (105); and a track score calculation unit (103) that calculates a track score in the current frame on the basis of a track score as an evaluation function for each resolution cell in the previous frame, the amplitude likelihood for the current frame calculated by the amplitude likelihood calculation unit (101), and the transition probability of the target from the previous frame to the current frame calculated by the transition probability calculation unit (102). The transition probability calculation unit (102) uses the tracking filter estimate calculated by the tracking processing unit (105) as a higher-order state quantity.

Description

Target detection device and target detection method

[0001] The present disclosure relates to a target detection device and a target detection method for detecting a target.

[0002] In multi-frame observations by any observation sensor, if the target signal in a single frame is weak, the signal strength of the target can be increased by adding signals across multiple frames, and randomly occurring signals such as noise can be canceled out. This makes it possible to enhance the target signal and make detection easier, even if the target signal in a single frame is weak.

[0003] However, in situations where the resolution cell in which the target exists varies significantly between frames due to target movement, adding signals from the same resolution cell across multiple frames will not add the target signal, and no enhancement effect of the target signal can be expected.

[0004] In contrast, conventional TBD (Track Before Detect) selects a resolution cell where the target exists for each frame, taking into account the movement of the target, and adds signals along the selected resolution cell. This makes it possible to enhance the weak signal of a moving target and enable detection, even in situations where the resolution cell where the target exists varies significantly between frames due to the target's movement.

[0005] On the other hand, in the conventional TBD, the transition probability constituting the track score, which is an evaluation function, is set on the assumption that the target motion is roughly constant. Therefore, the conventional TBD has a problem in that the detection accuracy deteriorates when the target motion changes.

[0006] In response to this, ATS-TBD (Adaptive Transition Set-TBD) has been proposed, which solves this problem by adaptively controlling the transition probability through motion estimation using a tracking filter (see, for example, Non-Patent Document 1).

[0007] X. Li, et al., “A DP-TBD Algorithm with Adaptive State Transition Set for Maneuvering Targets”, 2016 CIE International Conference on Radar (RADAR)

[0008] Here, the track score, which is an evaluation function of TBD, is expressed as the sum of the signal amplitude likelihood and the transition probability that evaluates the likelihood of the transition of the resolution cell between frames. In TBD, it is necessary to update the track score for each resolution cell in the sensor image for each frame using the following formula (1). In formula (1), f represents the frame number, and S ｆ indicates the track score, and x ｆ denotes the resolution cell, and L ｆ denotes the amplitude likelihood, and D(x ｆ ) is the resolution cell (x ｆ ) transition source search range, and P(x ｆ |x ｆ－１ ) indicates that the target is located in the resolution cell (x) between frame (f-1) and frame (f). ｆ―１ ) to the resolution cell (x ｆ ) is shown.

[0009] Conventional ATS-TBD is configured to apply tracking filters independently to all resolution cells of a sensor image, and select a transition source that takes into account the target's motion and changes by using the statistical distance from the predicted value as the transition probability. This conventional ATS-TBD configuration has the problem of requiring a huge amount of calculation because it updates the tracking filters for the number of resolution cells per frame. Furthermore, the estimated and predicted values ​​of the tracking filters contain errors and do not necessarily represent the true value of the target position. However, directly incorporating these errors into the transition probability can result in the transition probability being biased toward the predicted values ​​of the tracking filters.

[0010] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a target detection device that can reduce the calculation load compared to conventional devices.

[0011] A target detection device according to the present disclosure includes an amplitude likelihood calculation unit that calculates an amplitude likelihood for each frame based on a plurality of consecutive frames; a tracking processing unit that calculates a tracking filter estimate by performing target tracking processing using N tracking filters, the number of which is smaller than the number of resolution cells in the frame; a transition probability calculation unit that calculates a target transition probability from the previous frame to the current frame based on the frame and the calculation results by the tracking processing unit; and a track score calculation unit that calculates a track score for the current frame based on a track score that is an evaluation function for each resolution cell in the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit, and the target transition probability from the previous frame to the current frame calculated by the transition probability calculation unit, and is characterized in that the transition probability calculation unit uses the tracking filter estimate calculated by the tracking processing unit as a high-order state quantity.

[0012] According to the present disclosure, the above-described configuration makes it possible to reduce the calculation load compared to the conventional technology.

[0013] FIG. 10A is a block diagram showing an example of the configuration of a target detection device according to embodiment 1. FIG. 10B is a flowchart showing an example of the operation of the target detection device according to embodiment 1. FIG. 10C is a block diagram showing an example of the configuration of a target detection device according to embodiment 2. FIG. 10D is a diagram showing an example of the operation of the target detection device according to embodiment 2. FIG. 10E is a block diagram showing an example of the configuration of a target detection device according to embodiment 3. FIG. 10F is a diagram showing an example of the operation of the target detection device according to embodiment 3. FIG. 10G is a block diagram showing an example of the configuration of a target detection device according to embodiment 4. FIG. 10H is a diagram showing an example of the operation of the target detection device according to embodiment 4. FIGS. 10A and 10B are diagrams showing examples of the hardware configuration of target detection devices according to embodiments 1 to 4.

[0014] Hereinafter, embodiments will be described in detail with reference to the drawings. Embodiment 1. Fig. 1 is a block diagram showing an example configuration of a target detection device 1 according to embodiment 1. In addition to the target detection device 1, Fig. 1 also shows an image acquisition unit 2 and a detection processing unit 3 connected to the target detection device 1.

[0015] The image acquisition unit 2 acquires a plurality of consecutive frames. Data representing the frames acquired by the image acquisition unit 2 is output to the target detection device 1.

[0016] The target detection device 1 performs TBD processing on a frame acquired by the image acquisition unit 2 using a track score (track score distribution). The track score is an evaluation function for each resolution cell in the frame. This allows the target detection device 1 to acquire an integral path that adds up target signals that are weak and difficult to detect in a single frame, and enhances and detects the target signals by adding signals along the acquired path. Data indicating the detection results by the target detection device 1 is output to the detection processing unit 3.

[0017] The detection processing unit 3 performs various processes based on the detection results from the target detection device 1.

[0018] As shown in FIG. 1, the target tracking device includes an amplitude likelihood calculation unit 101, a transition probability calculation unit 102, a track score calculation unit 103, a tracking input extraction unit 104, a tracking processing unit 105, and a gate processing unit 106.

[0019] The amplitude likelihood calculation unit 101 calculates the amplitude likelihood for each frame based on the frames acquired by the image acquisition unit 2. That is, the amplitude likelihood calculation unit 101 calculates the amplitude likelihood of the frame from the signal intensities of all resolution cells in the frame.

[0020] The transition probability calculation unit 102 calculates the transition probability of the target from the previous frame to the current frame based on the frame acquired by the image acquisition unit 2 and the tracking filter estimate value determined by the gate processing unit 106. That is, the transition probability calculation unit 102 calculates the transition probability from the resolution cell within the search range in the previous frame for all resolution cells in the current frame based on the tracking filter estimate value that is the result of the tracking processing and gate processing for the previous frame. In addition, at this time, the transition probability calculation unit 102 uses the tracking filter estimate value determined by the gate processing unit 106 as the higher-order state quantity and uses the observation specifications as the lower-order state quantity. Note that the transition probability calculation unit 102 does not perform processing on the first frame.

[0021] The track score calculation unit 103 calculates a track score for the current frame based on the track score calculated by the track score calculation unit 103 for the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit 101, and the target transition probability from the previous frame to the current frame calculated by the transition probability calculation unit 102. That is, for all resolution cells in the current frame, the track score calculation unit 103 first selects a transition source that maximizes the track score using equation (1) based on the target transition probability from the previous frame to the current frame and the track score for the previous frame. Then, the track score calculation unit 103 calculates the track score for the current frame by adding the value at the selected transition source to the amplitude likelihood of the current frame. Note that for the first frame, the track score calculation unit 103 uses the amplitude likelihood as the track score.

[0022] Based on the track score of the current frame calculated by the track score calculation unit 103, the tracking input extraction unit 104 extracts N track score peaks as a state vector (observation vector) to be input to a tracking filter used in the tracking processing unit 105. N is a number smaller than the number of resolution cells in the frame. In the following, the N track score peaks may also be referred to as tracking inputs.

[0023] The tracking processing unit 105 calculates a tracking filter estimate by performing target tracking processing using a tracking filter based on the extraction result by the tracking input extraction unit 104. That is, the tracking processing unit 105 performs target tracking processing by performing smoothing processing and prediction processing using the tracking filter, using the tracking input extracted by the tracking input extraction unit 104 as input. This tracking processing unit 105 performs tracking processing using N tracking filters, which is fewer than the number of resolution cells.

[0024] Based on the tracking filter estimate calculated by the tracking processing unit 105, the gate processing unit 106 determines the tracking filter estimate to be used in the transition probability calculation unit 102 by using an error ellipse obtained from the smoothed error covariance matrix associated with the tracking filter estimate.

[0025] Next, an example of the operation of the target detection device 1 according to the first embodiment shown in FIG. 1 will be described with reference to FIG.

[0026] In the example of operation of the target detection device 1 according to the first embodiment shown in FIG. 1, as shown in FIG. 2, the amplitude likelihood calculation unit 101 first calculates the amplitude likelihood for each frame based on the frames acquired by the image acquisition unit 2 (step ST101). That is, the amplitude likelihood calculation unit 101 calculates the amplitude likelihood (L ｆ (x ｆ )) to calculate.

[0027] The transition probability calculation unit 102 calculates the transition probability of the target from the previous frame to the current frame based on the frames acquired by the image acquisition unit 2 and the tracking filter estimated value determined by the gate processing unit 106 (step ST102). That is, the transition probability calculation unit 102 calculates the transition probability of the target from the previous frame (f-1) for all resolution cells in the current frame (f) based on the search range (D(x ｆ－１ )) in the resolution cell (x ｆ－１ ) from the transition probability (P(x ｆ |x ｆ－１)) based on the tracking filter estimated value that is the result of the tracking process and gate process for the previous frame (f-1). At this time, the transition probability calculation unit 102 uses the tracking filter estimated value determined by the gate processing unit 106 as the higher-order state quantity and uses the observation specifications as the lower-order state quantity. Note that the transition probability calculation unit 102 does not perform processing on the first frame.

[0028] Next, the track score calculation unit 103 calculates a track score for the current frame based on the track score calculated by the track score calculation unit 103 for the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit 101, and the transition probability of the target from the previous frame to the current frame calculated by the transition probability calculation unit 102 (step ST103). That is, first, the track score calculation unit 103 calculates the transition probability (P(x ｆ |x ｆ－１ )), and the track score (S ｆ－１ (x ｆ－１ )), the track score calculation unit 103 selects a transition source that maximizes the track score using equation (1). Then, the track score calculation unit 103 calculates the value at the selected transition source as the amplitude likelihood (L ｆ (x ｆ )) to obtain the track score (S ｆ (x ｆ ) for the first frame. １ (x １ )) to track score (S １ (x １ ))

[0029] Next, the tracking input extraction unit 104 extracts N track score peaks as state vectors (observation vectors) to be input to the tracking filter used in the tracking processing unit 105 based on the track score of the current frame calculated by the track score calculation unit 103 (step ST104).

[0030] In this tracking input extraction unit 104, the number of observation vectors input to the tracking processing unit 105 is limited to N (<the number of resolution cells), thereby limiting the number of tracking processes performed in processing for each frame to N. This makes it possible to reduce the calculation load in the target detection device 1 according to embodiment 1 compared to when tracking processes are performed for the number of resolution cells in a frame.

[0031] The tracking input extraction unit 104 extracts N track score peaks, for example, by using the track score (S ｆ (x ｆ )), one method is to extract the first to Nth peak cells as the observation vector of the tracking input, based on the fact that high-score resolution cells are resolution cells in which the target is likely to exist.

[0032] Next, the tracking processing unit 105 calculates a tracking filter estimate value by performing target tracking processing using a tracking filter based on the extraction result by the tracking input extraction unit 104 (step ST105). That is, the tracking processing unit 105 performs target tracking processing by performing smoothing processing and prediction processing using the tracking filter, using the tracking input (N track score peaks) extracted by the tracking input extraction unit 104 as input. This tracking processing unit 105 performs tracking processing using N tracking filters, which is fewer than the number of resolution cells.

[0033] Here, the tracking filter can be an existing filter such as an α-β filter or a Kalman filter, and a description of the specific operation of the tracking filter will be omitted. The state vector of the tracking filter also includes a higher-order state quantity that does not appear in the observation vector, i.e., a state quantity of a dimension corresponding to the rate of change of motion. This allows the target detection device 1 according to the first embodiment to respond to changes in the motion of the target.

[0034] Next, the gate processing unit 106 determines a tracking filter estimated value to be used in the transition probability calculation unit 102, based on the tracking filter estimated value calculated by the tracking processing unit 105, using an error ellipse obtained from a smoothed error covariance matrix attached to the tracking filter estimated value (step ST106). That is, the gate processing unit 106 determines which of the first to Nth tracking filter estimated values ​​to use in the transition future prediction performed by the transition probability calculation unit 102, using the error ellipse.

[0035] Here, the gating process and subsequent transition probability calculation will be described using an example in which an image is two-dimensional data of distance and velocity, and a state vector with an acceleration term added as a higher-order state quantity is estimated using a tracking filter, as shown in FIG. 3 . Here, it is assumed that the search range (12 in FIG. 3 ) of the resolution cell of interest (11 in FIG. 3 ) of the current frame (f) for the previous frame (f−1) is included in the error ellipse (13 in FIG. 3 ) of the tracking filter (n (n = 1 to N)). Note that in the first embodiment, it is assumed that the error ellipses of the tracking filters do not overlap. In this case, the gating unit 106 links the resolution cell of interest (14 in FIG. 3 ) to the gate of the tracking filter (n) and outputs data indicating the state vector estimated by the tracking filter (n), along with a flag (in-gate flag) indicating that the resolution cell is within the gate of the tracking filter (n). On the other hand, if the search range is not included in the error ellipse of any of the tracking filters, the gate processing unit 106 links the search range to the resolution cell from which the transition originates and outputs data indicating a flag (outside-gate flag) indicating that the resolution cell is outside the gate of the tracking filter (n).

[0036] When the gate processing unit 106 outputs data indicating an intra-gate flag, the transition probability calculation unit 102 calculates the predicted transition destination (reference numeral 15 in FIG. 5) of the transition source resolution cell (reference numeral 14 in FIG. 5) one frame later using the following formula (2). In formula (2), x (hat) ｆ｜ｆ－１ indicates the predicted transition destination, Δt indicates the frame interval, and x ｆ－１ indicates a two-dimensional vector consisting of the coordinate values ​​of the distance and velocity of the resolution cell from which the transition originates, and a (hat) ｆ－１｜ｆ－１indicates the acceleration estimate (high-order state quantity) obtained by the tracking filter.

[0037] In this way, the transition probability calculation unit 102 uses only the high-order state quantity when predicting the future transition, without using all of the tracking filter estimated values. As a result, the target detection device 1 according to the first embodiment can avoid the track score being overweighted by the tracking predicted value.

[0038] The transition probability calculation unit 102 calculates the transition probability using the predicted transition destination and the resolution cell of interest, using the following equations (3) to (5). ｍａｘ denotes the maximum parameter of the transition probability, S denotes the residual covariance matrix, and R ｆ denotes the error between the predicted transition destination and the resolution cell of interest (reference numeral 16 in FIG. 3), and H denotes the observation matrix of the tracking filter.

[0039] On the other hand, when the gate processing unit 106 outputs data indicating an out-of-gate flag, the transition probability calculation unit 102 calculates the transition probability by processing similar to that of the conventional TBD, for example, based on the distance between the resolution cell of the transition source and the resolution cell of interest. Details of the transition probability calculation by the conventional TBD are omitted here.

[0040] In the above, we have shown a case where the observation parameters are two-dimensional data of distance and speed, such as when using radar. In contrast, when using a sensor such as an optical camera or an infrared sensor, distance and speed cannot be obtained, and only angle (direction) information can be obtained. However, the target detection device 1 can also be used in such cases.

[0041] For example, the observation parameters are two-dimensional data of the elevation angle and azimuth angle, and a state vector is estimated by adding an angular velocity term or the like as a higher-order state quantity using a tracking filter. In this case, the transition probability calculation unit 102 calculates the predicted transition destination using the following equation (6). In equation (6), x (hat) ｆ｜ｆ－１ indicates the predicted transition destination, Δt indicates the frame interval, and x ｆ－１ denotes a two-dimensional vector consisting of the coordinate values ​​of the elevation and azimuth angles of the resolution cell from which the transition originates, and v (hat)ｆ－１｜ｆ－１ indicates a two-dimensional vector consisting of the elevation and azimuth components of the angular velocity estimated by the tracking filter.

[0042] In this way, in the target detection device 1 according to embodiment 1, the tracking ability of a target whose motion changes is improved by using a tracking filter in the transition probability calculation, while the calculation load for each frame is reduced by limiting the number of tracking filters to less than the number of resolution cells, and further, by limiting the tracking filter estimated values ​​used in the transition probability calculation to only high-order state quantities, it is possible to avoid bias towards the tracking predicted value in the track score.

[0043] As described above, according to the first embodiment, the target detection device 1 includes the amplitude likelihood calculation unit 101 that calculates an amplitude likelihood for each frame based on a plurality of consecutive frames; the tracking processing unit 105 that calculates a tracking filter estimate value by performing target tracking processing using N tracking filters, the number of which is smaller than the number of resolution cells in the frame; the transition probability calculation unit 102 that calculates a transition probability of the target from the previous frame to the current frame based on the frame and the calculation result by the tracking processing unit 105; and the track score calculation unit 103 that calculates a track score for the current frame based on the track score, which is an evaluation function for each resolution cell in the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit 101, and the transition probability of the target from the previous frame to the current frame calculated by the transition probability calculation unit 102, and the transition probability of the target calculated by the transition probability calculation unit 102. The transition probability calculation unit 102 uses the tracking filter estimate value calculated by the tracking processing unit 105 as a high-order state quantity. The target detection device 1 according to the first embodiment also includes a tracking input extraction unit 104 that extracts, based on the track score calculated by the track score calculation unit 103, a tracking input that is N track score peaks as a state vector to be input to a tracking filter used in the tracking processing unit 105, and a gate processing unit 106 that determines, based on the tracking filter estimate calculated by the tracking processing unit 105, a tracking filter estimate used in the transition probability calculation unit 102 using an error ellipse obtained from a smoothed error covariance matrix associated with the tracking filter estimate. The transition probability calculation unit 102 uses the tracking filter estimate determined by the gate processing unit 106 as a higher-order state quantity and the observation parameters as a lower-order state quantity. According to the first embodiment, the transition probability calculation unit 102 calculates the target transition probability by setting the number of targets to two, using acceleration as the higher-order state quantity, and velocity as the lower-order state quantity. As a result, the target detection device 1 according to the first embodiment can reduce the computational load compared to conventional systems. Furthermore, the target detection device 1 according to the first embodiment can avoid biasing of the track score to the vicinity of the tracking predicted value in the transition source selection.

[0044] Furthermore, according to this embodiment 1, the target detection method includes the steps of: an amplitude likelihood calculation unit 101 calculating an amplitude likelihood for each frame based on a plurality of consecutive frames; a tracking processing unit 105 calculating a tracking filter estimate value by performing target tracking processing using N tracking filters, the number of which is smaller than the number of resolution cells in the frame; a transition probability calculation unit 102 calculating a transition probability of the target from the previous frame to the current frame based on the frames and the calculation results by the tracking processing unit 105; and a track score calculation unit 103 calculating a track score for the current frame based on the track score, which is an evaluation function for each resolution cell in the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit 101, and the transition probability of the target from the previous frame to the current frame calculated by the transition probability calculation unit 102, wherein the transition probability calculation unit 102 uses the tracking filter estimate value calculated by the tracking processing unit 105 as a high-order state quantity. As a result, the target detection method according to embodiment 1 can reduce the calculation load compared to conventional methods. Furthermore, in the target detection method according to the first embodiment, it is possible to avoid biasing of the track score to the vicinity of the tracking predicted value in the transition source selection.

[0045] Embodiment 2. In the first embodiment, target detection in a case where the error ellipses of the tracking filters do not overlap is described. In contrast, in the second embodiment, target detection in a case where the error ellipses of multiple tracking filters overlap is described. FIG. 4 is a block diagram showing an example configuration of a target detection device 1 according to the second embodiment. In the example configuration of the target detection device 1 according to the second embodiment shown in FIG. 4, the transition probability calculation unit 102 is changed to a transition probability calculation unit 102b in comparison with the example configuration of the target detection device 1 according to the first embodiment shown in FIG. 1. The other example configuration of the target detection device 1 according to the second embodiment shown in FIG. 4 is the same as the example configuration of the target detection device 1 according to the first embodiment, and only the differences will be described.

[0046] The transition probability calculation unit 102b has an additional function of performing target allocation processing compared to the transition probability calculation unit 102 in embodiment 1. In the function of performing target allocation processing possessed by the transition probability calculation unit 102b, when the gate processing unit 106 determines that multiple error ellipses overlap, that is, when the gate processing unit 106 outputs data indicating in-gate flags for multiple tracking filters, the function of performing target allocation processing selects a tracking filter estimate to be used from the tracking filter estimates determined by the gate processing unit 106 based on the track score calculated by the track score calculation unit 103. Then, the transition probability calculation unit 102b calculates the transition probability of the target using the tracking filter estimate selected by the function of performing target allocation processing.

[0047] Here, the target assignment process will be described in detail using an example in which the image is two-dimensional data of distance and velocity, and the search range (reference numeral 12 in FIG. 5 ) is included in the error ellipses (reference numerals 21 and 22 in FIG. 5 ) of the first and second tracking filters, as shown in FIG. 5 . In this case, the transition probability calculation unit 102b first calculates two predicted transition destinations (reference numerals 23 and 24 in FIG. 6 ) using the coordinate values ​​of the resolution cell (reference numeral 14 in FIG. 5 ) of the transition source and the acceleration estimates by the first and second tracking filters, respectively, according to equation (2). The transition probability calculation unit 102b then calculates the errors (reference numerals 25 and 26 in FIG. 6 ) between the resolution cell of interest and each of the two predicted transition destinations according to equations (4) and (5), and obtains the transition probability according to equation (3). The transition probability calculation unit 102b then compares the two obtained transition probabilities and selects the one that results in the larger track score.

[0048] Note that, through the target allocation process in the transition probability calculation unit 102b, one transition probability for each resolution cell of the transition source is input to the track score calculation unit 103. Therefore, similar to the first embodiment, the track score calculation unit 103 selects a transition source that maximizes the track score within the search area using equation (1) and calculates the track score.

[0049] In this way, according to the second embodiment, even in the case of proximity of targets, target scores can be processed separately without confusion by the target allocation process, making it possible to detect multiple proximity targets.

[0050] As described above, according to the second embodiment, when the gate processing unit 106 determines that multiple error ellipses overlap, the transition probability calculation unit 102b calculates the transition probability of the target after selecting a tracking filter estimated value to be used from the tracking filter estimated values ​​determined by the gate processing unit 106 based on the track score calculated by the track score calculation unit 103. This makes it possible for the target detection device 1 according to the second embodiment to process even cases where the error ellipses of multiple tracking filters overlap.

[0051] Embodiment 3. In the second embodiment, when the error ellipses of a plurality of tracking filters overlap, the transition probabilities are calculated for the number of overlapping tracking filters, and target allocation is performed based on a comparison of the magnitude of the track score. In contrast, in the third embodiment, a case will be described in which a function for performing target allocation is provided before the transition probability calculation is performed.

[0052] Fig. 6 is a block diagram showing an example configuration of a target detection device 1 according to embodiment 3. In the target detection device 1 according to embodiment 3 shown in Fig. 6, the transition probability calculation unit 102 is changed to a transition probability calculation unit 102c, and a target allocation unit 107 is added to the target detection device 1 according to embodiment 1 shown in Fig. 1. The other configuration examples of the target detection device 1 according to embodiment 3 shown in Fig. 6 are similar to the configuration example of the target detection device 1 according to embodiment 1, and the same reference numerals are used and only the different parts will be described.

[0053] When the gate processing unit 106 determines that multiple error ellipses overlap, that is, when the gate processing unit 106 outputs data indicating in-gate flags for multiple tracking filters, the target assignment unit 107 selects, based on the likelihood of the tracking filter, a tracking filter estimate to be used in the transition probability calculation unit 102c from the tracking filter estimates determined by the gate processing unit 106. Then, the transition probability calculation unit 102c calculates the transition probability of the target using the tracking filter estimate selected by the target assignment unit 107.

[0054] Here, the target assignment process will be described in detail using an example in which the image is two-dimensional data of distance and velocity, and the search range (reference numeral 12 in FIG. 7) is included in the error ellipses (reference numerals 21 and 22 in FIG. 7) of the first and second tracking filters, as shown in FIG. 7. In this case, the target assignment unit 107 first refers to the likelihood distributions (reference numerals 33 and 34 in FIG. 7) of the error ellipses (reference numerals 21 and 22 in FIG. 7) in which each resolution cell overlaps with the resolution cells (reference numerals 31, 32, and 35 in FIG. 7) of the source resolution cells. Then, as a result of the reference, the target allocation unit 107 outputs data indicating the in-gate flag and state vector of the first tracking filter to the resolution cell of the transition source (reference number 31 in FIG. 7) where the likelihood of the first tracking filter (reference number 33 in FIG. 7) is greater, and outputs data indicating the in-gate flag and state vector of the second tracking filter to the resolution cell of the transition source (reference number 32 in FIG. 7) where the likelihood of the second tracking filter (reference number 34 in FIG. 7) is greater, and so on, outputting data indicating the in-gate flag and state vector of the tracking filter with the higher likelihood. Note that the target allocation unit 107 outputs all of the data indicating the in-gate flag and state vector of the first and second tracking filters to the resolution cell of the transition source (reference number 35 in FIG. 7) where the likelihoods of the first and second tracking filters are equal.

[0055] Then, for a transition source resolution cell associated with only one intra-gate flag and state vector by the target assignment unit 107, the transition probability calculation unit 102c calculates a predicted transition destination (reference numerals 36 and 37 in FIG. 7 ) using equation (2) and calculates transition probabilities using equations (3) to (5), similarly to the transition probability calculation unit 102 in embodiment 1. On the other hand, for a transition source resolution cell (reference numeral 35 in FIG. 7 ) output by the target assignment unit 107 while holding two intra-gate flags and state vectors, the transition probability calculation unit 102c performs target assignment in the same manner as the transition probability calculation unit 102b in embodiment 2, calculating two predicted transition destinations (reference numerals 38 and 39 in FIG. 7 ) using the estimated values ​​of the first and second tracking filters using equations (4) and (5), calculating two transition probabilities using equation (3), and selecting a tracking filter that increases the track score.

[0056] Note that one transition probability for each resolution cell of the transition source is input to the track score calculation unit 103. Therefore, similar to the first embodiment, the track score calculation unit 103 selects a transition source that maximizes the track score within the search area using equation (1) to calculate the track score.

[0057] As described above, according to this embodiment 3, even when a target is approaching, the target scores can be processed separately without confusion by the target allocation process, making it possible to detect multiple nearby targets, and also reducing the number of transition probability calculations by referring to the likelihood of each tracking filter, thereby reducing the computational load.

[0058] As described above, according to this embodiment 3, the target detection device 1 includes a target assignment unit 107 that, when it is determined by the gate processing unit 106 that a plurality of error ellipses overlap, selects a tracking filter estimate to be used by the transition probability calculation unit 102c from the tracking filter estimates determined by the gate processing unit 106 based on the likelihood of the tracking filter, and the transition probability calculation unit 102c calculates the transition probability of the target using the tracking filter estimate selected by the target assignment unit 107. As a result, the target detection device 1 according to embodiment 3 can process even cases where the error ellipses of a plurality of tracking filters overlap, and can reduce the calculation load.

[0059] Embodiment 4. In the first embodiment, a case was described in which a high-order state quantity, which is a rate of change of motion, is acquired by a tracking process performed for each frame, and a predicted transition destination calculation and a transition probability calculation are performed. In contrast, in the fourth embodiment, a case will be described in which a tracking input is recorded, and a function is provided to estimate a high-order state quantity by referring to and re-smoothing a plurality of tracking inputs.

[0060] Fig. 8 is a block diagram showing an example configuration of a target detection device 1 according to embodiment 4. In the target detection device 1 according to embodiment 4 shown in Fig. 8, a tracking input recording unit 108 and a tracking input smoothing processing unit 109 are added to the target detection device 1 according to embodiment 1 shown in Fig. 1. The other configuration examples of the target detection device 1 according to embodiment 4 shown in Fig. 8 are similar to the configuration example of the target detection device 1 according to embodiment 1, and the same reference numerals are used, and only the different parts will be described.

[0061] The tracking input recording unit 108 records data indicating the tracking inputs (N track score peaks) extracted by the tracking input extraction unit 104 for each frame.

[0062] The tracking input recording unit 108 may be, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), or an EEPROM (Electrically EPROM), a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc).

[0063] 8 shows a case where the tracking input recording unit 108 is provided inside the target detection device 1. However, the present invention is not limited to this, and the tracking input recording unit 108 may be provided outside the target detection device 1.

[0064] Tracking input smoothing unit 109 obtains a high-order state quantity of the target by performing smoothing processing on the tracking input for multiple frames based on the tracking input for multiple frames extracted by tracking input extraction unit 104. At this time, tracking input smoothing unit 109 extracts the tracking input for multiple frames from the data recorded in tracking input recording unit 108 and performs smoothing processing on the tracking input.

[0065] Furthermore, the transition probability calculation unit 102 calculates the transition probability of the target from the previous frame to the current frame based on the frames acquired by the image acquisition unit 2 and the results acquired by the tracking input smoothing processing unit 109 .

[0066] Here, an example of the operation of the tracking input smoothing processor 109 will be described using an example in which tracking inputs in three consecutive frames are extracted and smoothed, as shown in FIG. 9 . The tracking input smoothing processor 109 first acquires a resolution cell (42 in FIG. 9 ) from which a transition occurs within the search range and the associated past tracking inputs (43 and 44 in FIG. 9 ) for a resolution cell of interest in the current frame (f) (reference numeral 41 in FIG. 9 ). Next, the tracking input smoothing processor 109 considers the tracking inputs in these past frames to be past trajectories and smooths them, for example, by polynomial fitting (reference numeral 45 in FIG. 9 ). The tracking input smoothing processor 109 then outputs data indicating estimates, which are high-order state quantities of the target obtained through this smoothing process.

[0067] Furthermore, the transition probability calculation unit 102 performs transition prediction using the high-order state quantities obtained by the tracking input smoothing processing unit 109. As a result, the target detection device 1 according to embodiment 4 can obtain a predicted transition destination (reference numeral 47 in FIG. 8) that is more in line with the movement of the target, compared to a predicted transition destination (reference numeral 46 in FIG. 8) that uses a tracking filter estimated value for the state quantity of one frame before.

[0068] As described above, according to the fourth embodiment, by recording data indicating the tracking input and smoothing the tracking input for a plurality of frames, it is possible to reduce the influence of estimation errors caused by a normal tracking filter and to output data indicating more stable high-order state quantity estimation values.

[0069] In the above description, the tracking input recording unit 108 and the tracking input smoothing processing unit 109 are added to the target detection device 1 according to embodiment 1. However, the present invention is not limited to this, and the tracking input recording unit 108 and the tracking input smoothing processing unit 109 may be added to the target detection device 1 according to embodiments 2 and 3, and the same effects as those described above can be obtained.

[0070] As described above, according to this fourth embodiment, the target detection device 1 includes a tracking input smoothing unit 109 that obtains a higher-order state quantity of the target by performing smoothing processing on the tracking input for multiple frames based on the tracking input for multiple frames extracted by the tracking input extraction unit 104, and the transition probability calculation unit 102 calculates the transition probability of the target from the previous frame to the current frame based on the frame and the results obtained by the tracking input smoothing unit 109. This makes it possible to further stabilize the transition probability in the target detection device 1 according to the fourth embodiment.

[0071] Finally, with reference to FIG. 10 , an example of the hardware configuration of the target detection device 1 according to the first to fourth embodiments will be described. Note that, although an example of the hardware configuration of the target detection device 1 according to the first embodiment will be described below, the same applies to the example of the hardware configuration of the target detection device 1 according to the second to fourth embodiments. The functions of the amplitude likelihood calculation unit 101, the transition probability calculation unit 102, the track score calculation unit 103, the tracking input extraction unit 104, the tracking processing unit 105, and the gate processing unit 106 in the target detection device 1 are realized by a processing circuit 51. The processing circuit 51 may be dedicated hardware as shown in FIG. 10A , or may be a CPU (also referred to as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)) 52 that executes a program stored in a memory 53 as shown in FIG. 10B .

[0072] When the processing circuit 51 is dedicated hardware, the processing circuit 51 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The functions of the amplitude likelihood calculation unit 101, the transition probability calculation unit 102, the track score calculation unit 103, the tracking input extraction unit 104, the tracking processing unit 105, and the gate processing unit 106 may be realized individually by the processing circuit 51, or the functions of each unit may be realized collectively by the processing circuit 51.

[0073] When the processing circuit 51 is a CPU 52, the functions of the amplitude likelihood calculation unit 101, transition probability calculation unit 102, track score calculation unit 103, tracking input extraction unit 104, tracking processing unit 105, and gate processing unit 106 are realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in memory 53. The processing circuit 51 realizes the functions of each unit by reading and executing the programs stored in memory 53. In other words, the target detection device 1 includes memory 53 for storing programs that, when executed by the processing circuit 51, result in the execution of, for example, each step shown in FIG. 2 . Furthermore, these programs can also be said to cause a computer to execute the procedures and methods of the amplitude likelihood calculation unit 101, transition probability calculation unit 102, track score calculation unit 103, tracking input extraction unit 104, tracking processing unit 105, and gate processing unit 106. Here, the memory 53 may be, for example, a non-volatile or volatile semiconductor memory such as a RAM, a ROM, a flash memory, an EPROM, or an EEPROM, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD.

[0074] It should be noted that the functions of the amplitude likelihood calculation unit 101, the transition probability calculation unit 102, the track score calculation unit 103, the tracking input extraction unit 104, the tracking processing unit 105, and the gate processing unit 106 may be partially implemented by dedicated hardware and partially implemented by software or firmware. For example, the function of the amplitude likelihood calculation unit 101 may be implemented by the processing circuit 51 as dedicated hardware, and the functions of the transition probability calculation unit 102, the track score calculation unit 103, the tracking input extraction unit 104, the tracking processing unit 105, and the gate processing unit 106 may be implemented by the processing circuit 51 reading out and executing programs stored in the memory 53.

[0075] In this way, the processing circuitry 51 can realize each of the above-described functions by hardware, software, firmware, or a combination of these.

[0076] It should be noted that the embodiments may be freely combined, or any of the components in each embodiment may be modified, or any of the components in each embodiment may be omitted.

[0077] The target detection device according to the present disclosure can reduce the calculation load compared to the conventional device, and is suitable for use in a target detection device that detects targets.

[0078] 1 Target detection device, 2 Image acquisition unit, 3 Detection processing unit, 51 Processing circuit, 52 CPU, 53 Memory, 101 Amplitude likelihood calculation unit, 102, 102b, 102c Transition probability calculation unit, 103 Track score calculation unit, 104 Tracking input extraction unit, 105 Tracking processing unit, 106 Gate processing unit, 107 Target allocation unit, 108 Tracking input recording unit, 109 Tracking input smoothing processing unit.

Claims

1. An amplitude likelihood calculation unit calculates the amplitude likelihood for each frame based on a series of consecutive frames, A tracking processing unit calculates a tracking filter estimate by performing target tracking processing using N tracking filters, which is fewer than the number of resolution cells in the aforementioned frame. A transition probability calculation unit calculates the probability of the target transitioning from the previous frame to the current frame based on the frame and the calculation results from the tracking processing unit, The system includes a track score calculation unit that calculates the track score for the current frame based on the track score, which is an evaluation function for each resolution cell in the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit, and the target transition probability from the previous frame to the current frame calculated by the transition probability calculation unit. The transition probability calculation unit uses the tracking filter estimate calculated by the tracking processing unit as a higher-order state variable. A target detection device characterized by the following features.

2. A tracking input extraction unit extracts N tracking inputs, which are track score peaks, as state vectors to be input to the tracking filter used in the tracking processing unit, based on the track score calculated by the track score calculation unit. The system includes a gate processing unit that determines the tracking filter estimate to be used in the transition probability calculation unit, based on the tracking filter estimate calculated by the tracking processing unit, using the error ellipse obtained from the smoothed error covariance matrix associated with the tracking filter estimate. The transition probability calculation unit uses the tracking filter estimate determined by the gate processing unit as the higher-order state quantity and the observed parameters as the lower-order state quantity. The target detection device according to claim 1, characterized in that it is a target detection device.

3. When the gate processing unit determines that multiple error ellipses overlap, the transition probability calculation unit, based on the track score calculated by the track score calculation unit, selects the tracking filter estimate to be used from among the tracking filter estimates determined by the gate processing unit, and then calculates the transition probability of the target. The target detection device according to claim 2, characterized in that it is a target detection device.

4. When the gate processing unit determines that multiple error ellipses overlap, the target assignment unit selects a tracking filter estimate to be used in the transition probability calculation unit from among the tracking filter estimates determined by the gate processing unit, based on the likelihood of the tracking filter. The transition probability calculation unit calculates the transition probability of the target using the tracking filter estimate selected by the target assignment unit. The target detection device according to claim 2, characterized in that it is a target detection device.

5. The system includes a tracking input smoothing processing unit that obtains a target higher-order state quantity by performing smoothing processing on multiple frames of tracking inputs extracted by the aforementioned tracking input extraction unit. The transition probability calculation unit calculates the target transition probability from the previous frame to the current frame based on the frame and the results obtained by the tracking input smoothing unit. A target detection device according to any one of claims 2 to 4, characterized by the features described above.

6. The transition probability calculation unit calculates the transition probability of the targets using equations (2) to (5), assuming there are two targets, a higher-order state quantity as acceleration, and a lower-order state quantity as velocity. A target detection device according to any one of claims 1 to 4, characterized in that

7. The amplitude likelihood calculation unit calculates the amplitude likelihood for each frame based on a plurality of consecutive frames. The tracking processing unit calculates an estimated tracking filter value by performing tracking processing on the target using N tracking filters, which is fewer than the number of resolution cells in the frame. The transition probability calculation unit calculates the transition probability of the target from the previous frame to the current frame based on the frame and the calculation results from the tracking processing unit. The track score calculation unit includes the step of calculating the track score for the current frame based on the track score, which is an evaluation function for each resolution cell in the previous frame, the amplitude likelihood of the current frame calculated by the amplitude likelihood calculation unit, and the target transition probability from the previous frame to the current frame calculated by the transition probability calculation unit. The transition probability calculation unit uses the tracking filter estimate calculated by the tracking processing unit as a higher-order state variable. A target detection method characterized by the following:

Images