JP2015064229A - Tracking device, tracking method, and tracking program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a moving object in an area observed separately in time while being divided into a plurality of sub-areas to be tracked with higher accuracy than in the past.SOLUTION: A tracking device according to the present invention comprises: a prediction unit which, every time each of a plurality of sub-areas created by dividing a prescribed area including a moving object is sequentially photographed separately in time, finds a gate area including a predicted position of the moving object on the basis of a track of the moving object estimated before a time of day at which the sub-area was photographed; an extraction unit for extracting, from each of the images of the plurality of sub-areas obtained by photographing, an observation point indicating a location included in the gate area found by the prediction unit and having a feature amount greater than or equal to a first threshold; an estimation unit for estimating, every time each of the plurality of sub-areas is photographed, a new track using the observation point extracted by the extraction unit and the track used in finding the gate area including the observation point; and a tracking unit for selecting, from among a plurality of tracks of the moving object including the new track estimated by the estimation unit, a track having higher likelihood than other tracks as a plausible track of the moving object.

Description

  The present invention relates to a tracking device, a tracking method, and a tracking program for tracking a moving object.

  Tracking devices that apply multiple hypothesis tracking (MHT) technology have been proposed to track the track of moving objects such as aircraft based on observation data obtained from radars and infrared sensors (for example, Non-Patent Document 1).

  The tracking device to which the MHT is applied considers the observation point and the moving object extracted from the observation data as a position having the possibility of the position of the moving object in consideration of the failure of detection of the moving object or the possibility of noise generation. Make multiple hypotheses about the correspondence between In addition, the tracking device to which MHT is applied tracks the track of the moving object by proceeding with the tracking process while evaluating the probability of each hypothesis.

  In a tracking device to which MHT is applied, it is assumed that an area including a moving object is shot at a predetermined time interval. In the tracking device to which the MHT is applied, a gate region including the predicted position of the moving object is obtained based on the past track of the moving object obtained by the tracking process based on the observation data captured so far. Used to extract observation points from new observation data.

  By the way, the tracking device captures an image of the entire observation area as each image obtained by sequentially capturing each of a plurality of partial areas obtained by dividing the observation area to be observed at a predetermined time interval. There are cases where the information is acquired at predetermined time intervals. This type of tracking device, for example, acquires images of each partial area included in the observation area sequentially by performing imaging while changing the direction of the infrared sensor, and acquires each time the imaging of each partial area is completed. By synthesizing the obtained images, an image of one frame corresponding to the entire observation region is acquired. Then, the tracking device estimates the past track of the moving object based on the images of several frames acquired so far, and uses the state vector indicating the position, velocity, acceleration, etc. of the moving object obtained by the estimation. Then, track the track of the moving object.

  In this type of tracking device, a gate area including a predicted position of the moving object is set based on the track of the moving object up to the previous frame at the time at which each partial area is imaged. Techniques used for tracking processing have been proposed (see, for example, Patent Document 1).

Japanese Patent Application No. 2004-233136

D. B. Reid, “An algorithm for tracking multiple targets.” IEEE Trans. On Automatic Control, vol.24, No.6, pp.843-854, 1979

  By the way, when the observation area is divided into a plurality of partial areas and photographed, the gate area predicted based on the track of the moving body up to the previous frame may extend over two or more partial areas. In this case, there is a possibility that one moving body may be observed in each of two or more partial areas while imaging of each partial area included in the observation area is completed.

  However, in the conventional technique, it is difficult to perform the tracking process in consideration of the possibility that one moving body is observed in each of two or more partial regions. For this reason, for example, when the speed of the moving body changes during the shooting of each partial area included in the observation area, the wake that has changed according to the change in speed may not be tracked and may be lost. is there.

  The tracking device, the tracking method, and the tracking program disclosed herein provide a technique that enables tracking of a moving object in a region observed in a time-division divided into a plurality of partial regions with higher accuracy than in the past. With the goal.

  According to one aspect, the tracking device is configured to detect a time before each time a partial area is captured each time a plurality of partial areas obtained by dividing a predetermined area including a moving object are sequentially captured in time division. Based on the estimated track of the moving body, each of the prediction unit that obtains the gate area including the predicted position of the moving body, and each of the images of the plurality of partial areas obtained by the imaging, is obtained by the prediction unit. An extraction unit for extracting an observation point indicating a portion having a feature value equal to or greater than the first threshold included in the gate region, and an observation point and an observation point extracted by the extraction unit each time a plurality of partial regions are photographed An estimator that estimates a new wake using the wake used to obtain the gate area including the wake, and a plurality of wakes of the moving object including the new wake estimated by the estimator. Higher likelihood than wake And a tracking unit for selecting track with a plausible track mobile.

  According to another aspect, the tracking method is performed before each time a partial area is captured each time a plurality of partial areas obtained by dividing a predetermined area including a moving object are sequentially captured in time division. Based on the estimated track of the moving body, each of the gate areas including the predicted position of the moving body is obtained, and the first threshold value included in the gate area is determined from each of the plurality of partial area images obtained by imaging. Each time a plurality of partial areas are photographed, the observation points extracted from the gate area and the wake used to determine the gate area are extracted. A new wake is estimated, and a wake having a higher likelihood than the other wakes is selected as a likely wake of the moving body from a plurality of wakes of the moving body including the estimated new wake.

  Further, according to another aspect, the tracking program is configured so that each time a plurality of partial areas obtained by dividing a predetermined area including a moving object are sequentially photographed in time division, the time at which the partial area is photographed. Based on the previously estimated track of the moving body, each of the gate areas including the predicted position of the moving body is obtained, and the first included in the gate area is obtained from each of the plurality of partial area images obtained by imaging. The observation points indicating the points having the feature amount equal to or greater than the threshold are extracted, and each time a plurality of partial areas are photographed, the observation points extracted from the gate area and the wake used to obtain the gate area are obtained. A new wake is estimated, and a wake having a higher likelihood than the other wakes is selected as a likely wake of the moving body from a plurality of wakes of the moving body including the estimated new wake. Compu processing To be executed by the data.

  The tracking device, the tracking method, and the tracking program disclosed in the present disclosure can track the track of a moving body in an area observed in a time-division divided into a plurality of partial areas with higher accuracy than in the past.

It is a figure which shows one Embodiment of a tracking apparatus. It is a figure which shows the example of a 1st track and a 2nd track. It is a figure which shows the example of a hypothesis tree. It is a figure which shows operation | movement of the tracking apparatus shown in FIG. It is a figure which shows another embodiment of a tracking apparatus. It is a figure which shows the example of calculation of the likelihood of each hypothesis shown in FIG. It is a figure which shows the example of the range which integrates the probability density of presence of a moving body in Formula (1). It is a figure which shows the example of the range which integrates the probability density of presence of a moving body in Formula (1). It is a figure which shows the example of the range which integrates the probability density of presence of a moving body in Formula (1). It is a figure which shows an example of the hardware constitutions of the tracking apparatus shown in FIG. It is a figure which shows the tracking process by the tracking apparatus shown in FIG.

  Hereinafter, embodiments will be described with reference to the drawings.

  FIG. 1 shows an embodiment of a tracking device. The tracking device 10 shown in FIG. 1 uses the imaging device EQ to detect the image in the observation region based on an image obtained by imaging a plurality of partial regions included in the observation region Ar that is an observation target region in a time division manner. Track the track of a moving object.

  The imaging device EQ includes, for example, an infrared sensor SC including pixels that are two-dimensionally arranged. In addition, the imaging device EQ rotates the light receiving surface of the infrared sensor SC in the horizontal directions Dh1, Dh2 and the vertical directions Dv1, Dv2. Thereby, the imaging device EQ sequentially captures 2m (m is a positive integer) partial areas dA_1,..., DA_m, dA_m + 1,..., DA_2m included in the observation area Ar using the infrared sensor SC. The 2m partial areas dA_1 to dA_2m shown in FIG. 1 divide the observation area Ar into two in the y-axis direction corresponding to the vertical direction, and to m in the x-axis direction corresponding to the horizontal direction. Although the case of dividing is shown, the number of divisions in the y-axis direction is not limited to two.

  For example, the imaging device EQ sequentially images m partial areas dA_1 to dA_m while rotating the direction of the infrared sensor SC in the horizontal direction Dh1 corresponding to the x-axis direction. Thereafter, the imaging apparatus EQ sequentially images the m partial areas dA_m + 1 to dA_2m while rotating the infrared sensor SC in the vertical direction Dv1 corresponding to the y-axis direction and then rotating in the horizontal direction Dh2. When the imaging of the partial area dA_2m is completed, that is, after the imaging of the 2m partial areas dA_1 to dA_2m included in the observation area Ar is completed, the imaging apparatus EQ changes the direction of the infrared sensor SC in the vertical direction Dv2. By rotating it, it is directed in the direction of the partial area dA_1. Then, the imaging apparatus EQ newly starts imaging of the partial areas dA_1 to dA_2m. That is, the imaging device EQ repeatedly acquires a frame image corresponding to the observation area Ar at a time interval corresponding to a repetition cycle by cyclically and repeatedly imaging the 2m partial areas dA_1 to dA_2m. For example, when the observation area Ar includes 54 (m = 27) partial areas, and the time required for imaging each partial area is about 0.2 seconds, the imaging apparatus EQ corresponds to the observation area Ar. The time interval for acquiring the image of one frame to be acquired is about a dozen seconds.

  In the following description, when the partial areas dA_1 to dA_2m are not distinguished from each other, the partial areas dA_1 to dA_2m are simply referred to as partial areas dA.

  The image processing unit GP included in the imaging device EQ has a predetermined first threshold value or more in each partial area dA based on the output signal of the infrared sensor SC obtained when the infrared sensor SC is directed to each partial area dA. Detects a spot with a brightness of. In addition, the image processing unit GP passes information indicating the position of each detected location to the tracking device 10 as an observation point indicating the position of an object that may be a moving object. Note that the image processing unit GP may be included in the tracking device 10.

  The tracking device 10 includes a prediction unit 11, an extraction unit 12, an estimation unit 13, and a tracking unit 14. The prediction unit 11 includes a range including a predicted position of the moving body based on the track of the moving body estimated by the estimation unit 13 every time each of the partial areas dA is captured by the imaging device EQ. Find the gate region. The extraction unit 12 receives an observation point detected by the image processing unit GP from an image obtained by shooting every time the partial area dA is shot, and is obtained by the prediction unit 11 from the received observation points. Extract observation points included in the gate area. Then, every time the partial area dA is photographed, the estimation unit 13 uses the observation points extracted by the extraction unit 12 and the wake used when obtaining the gate area to create a first wake that is a new wake. presume. The wake estimated by the estimation unit 13 is sequentially transferred to the prediction unit 11 and used for obtaining a gate region used for extraction of observation points from the partial region dA to be photographed thereafter. For example, the tracking unit 14 evaluates the likelihood of each wake estimated by the estimation unit 13 in the process of taking a picture of the partial area dA included in the observation area Ar, and includes at least one hypothesis including the highest likelihood hypothesis. Tracking a moving object by selecting one wake. The track selected by the tracking unit 14 is passed to the prediction unit 11, and as information indicating the past track of the moving body, the prediction of the gate region at the time when each of the partial regions dA to be newly imaged is imaged. Used for.

  For example, when the imaging unit EQ starts imaging of the observation area Ar in the k (k is an integer greater than 2) time, the mobile unit obtained by the tracking process up to the (k-1) th imaging is performed. The gate area at the photographing time of the partial area dA_1 is obtained based on the past wake. For example, the predicting unit 11 may observe the moving object at the time when the partial region dA_1 is observed when the moving object moves according to the past wake obtained by the tracking process up to the (k-1) th imaging. Is obtained as a gate area at the photographing time of the partial area dA_1. In addition, every time each subsequent partial area dA (for example, partial area dA_2) is captured, the prediction unit 11 captures each partial area dA based on each of the wakes estimated by the estimation unit 13 so far. The gate area at the specified time is obtained. Here, the motion of the moving body according to each estimated wake is indicated by a state vector including, for example, the position of the moving body, the speed of the moving body, and the acceleration of the moving body estimated for a past time. The predicting unit 11 obtains a gate region as a range in which the probability density of the presence of a moving object at a time when each partial region dA is observed by passing a state vector indicating each wake through a Kalman filter. In the following description, when the moving body moves in the wake indicated by each hypothesis, the position of the moving body predicted for the time at which each partial area dA is imaged is referred to as the predicted position of the moving body. . On the other hand, the position of the moving body estimated for a certain past time is referred to as the estimated position of the moving body.

  The estimation unit 13 generates a hypothesis that the moving body is observed at the extracted observation point when the extraction unit 12 extracts the observation point included in the gate region, and based on the generated hypothesis, the observation point The first wake is estimated using the information and the wake used for the prediction of the gate area. For example, the estimation unit 13 updates the state vector representing the wake used for the prediction of the gate region with the information on the observation point, so that the state vector indicating the first wake and the position of the observation point in the past are updated. A state vector reflecting both the wake and the wake is obtained.

  For example, when the observation point in the partial region dA photographed at the first time is extracted by the extraction unit 12, the estimation unit 13 indicates that the moving body is observed at the observation point in the partial region dA at the first time. Generate a hypothesis to And the estimation part 13 estimates the state vector which shows the 1st wake shown by the produced | generated hypothesis based on the information of an observation point, and the state vector which shows the past wake. The first wake indicated by the estimated state vector is a wake that passes through a position closer to the observation point than the predicted position of the mobile object predicted for the first time based on the past wake.

  Thereafter, the estimation unit 13 extracts the observation points extracted from the image of the partial area dA photographed at the second time using the gate area obtained from the first wake estimated at the time of photographing at the first time. And a new first wake is estimated based on the estimated first wake. For example, the estimation unit 13 assumes a hypothesis that the moving object is observed at both the observation point included in the partial area captured at the first time and the observation point included in the partial area captured at the second time. Generate. Then, the estimation unit 13 generates based on the state vector representing the first wake estimated at the time of shooting at the first time and the observation point information extracted from the image shot at the second time. A new state vector indicating a new first wake indicated by the hypothesis is estimated.

  Further, in addition to the first track described above, the estimation unit 13 does not use the observation point information, but based on the track used to obtain the gate region, the second track that is a new track is obtained. It is desirable to estimate. For example, the estimation unit 13 generates a hypothesis that no moving object is observed in the gate region used for extraction of observation points. Further, the generation unit 13 uses the predicted position indicated by the gate area, for example, and updates the estimated position included in the state vector indicating the track used for the prediction of the gate area, thereby indicating the generated hypothesis. A state vector indicating the second wake is estimated. The second wake indicated by the estimated state vector is a wake obtained by extending the wake used for obtaining the gate area.

  FIG. 2 shows an example of the first wake and the second wake. Each of the rectangles shown in FIG. 2 is obtained by the infrared sensor SC for the partial areas dA_1, dA_2, dA_2m-1, and dA_2m in the process of imaging the observation area Ar for the kth time by the imaging apparatus EQ shown in FIG. Shows the image. In FIG. 2, an image obtained by photographing the partial areas dA other than the partial areas dA_1, dA_2, dA_2m-1, and dA_2m among the partial areas dA_1 to dA_2m illustrated in FIG. 1 is omitted. ing.

  The solid line curve Trp shown in FIG. 2 is obtained by tracking the moving object performed in the process of observing the observation area Ar k−1 times before the k-th imaging of the observation area Ar. The past wake obtained as a reasonable wake is shown. Further, in the image dA_1 shown in FIG. 2, the point Q (Trp, k−1) and the point Q (Trp, k−2) are times when the partial area dA_1 is photographed at the (k−1) th time and the (k−2) th time. The estimated position which is the position of the moving body estimated based on the wake Trp is shown.

  In addition, an ellipse Gp (2) indicated by a solid line in FIG. 2 indicates the time at which the partial area dA_2 was imaged in the process of imaging the observation area Ar for the kth time based on the past track Trp of the moving object. 2 shows the gate region obtained by the prediction unit 11 shown in FIG. Further, in the partial area dA_2 shown in FIG. 2, a point Z1 indicates an example of an observation point detected by the image processing unit GP shown in FIG. 1 from an image obtained by photographing the partial area dA_2.

  In the example of FIG. 2, the observation point Z1 detected in the partial region dA_2 is included in the gate region Gp (2). Therefore, the extraction unit 12 illustrated in FIG. 1 passes information indicating the observation point Z1 to the estimation unit 13. The estimation unit 13 generates a hypothesis that the moving body is not observed in the gate region Gp (2) together with the hypothesis that the moving body is observed at the observation point Z1 received from the extraction unit 12.

  Then, the estimation unit 13 obtains a new state vector based on the state vector representing the past wake Trp and the information on the observation point Z1, thereby indicating the first hypothesis that the moving body is observed at the observation point Z1. Estimate one track. A curve Tr1 indicated by a broken line in FIG. 2 is an example of a wake estimated by the estimation unit 13 using the information of the observation point Z1 as the first wake. Further, the estimation unit 13 uses the wake Trp shown in FIG. 2 as the second wake indicated by the hypothesis that the moving body is not observed in the gate region Gp (2) based on the state vector representing the past wake Trp. Estimate the wake that extends.

  Thereafter, when the partial area dA_2m-1 is photographed, the prediction unit 11 obtains the gate area Gp (2m-1) surrounded by a dotted line in FIG. 2 based on the past track Trp of the moving body. Similarly, the prediction unit 11 encloses the time when the partial area dA_2m-1 is captured based on the wake Tr1 estimated as the first wake by the estimation unit 13 when the partial area dA_2 is captured, with a one-dot chain line. The gate region G1 (2m−1) shown in FIG.

  The state vector of the wake Tr1 used for calculation of the gate region G1 (2m−1) shown in FIG. 2 is obtained by using the observation point Z1 extracted from the partial region dA_2 when the partial region dA_2 is photographed. It is calculated | required by updating the state vector showing wake Trp. Therefore, as shown in FIG. 2, the position of the gate region G1 (2m-1) obtained based on the wake Tr1 is shifted from the gate region Gp (2m-1) obtained based on the past wake Trp. It becomes the position.

  The point Z2 included in the partial area dA_2m-1 illustrated in FIG. 2 is an example of an observation point detected by the image processing unit GP illustrated in FIG. 1 from an image obtained by photographing the partial area dA_2m-1. It is. In the example of FIG. 2, the observation point Z2 detected in the partial region dA_2m-1 is included in the gate region G1 (2m-1) predicted based on the wake Tr1. Accordingly, the extraction unit 12 illustrated in FIG. 1 passes the observation point Z2 to the estimation unit 13 as an observation point extracted from the gate region G1 (2m−1) predicted from the first track Tr1. On the other hand, the observation point Z2 shown in FIG. 2 is not included in the gate region Gp (2m−1) predicted based on the past wake Trp. Therefore, the extraction unit 12 does not pass the observation point Z2 to the estimation unit 13 as an observation point in the gate region Gp (2m−1) predicted based on the past wake Trp.

  In this case, the estimation unit 13 assumes that a moving body is observed at the observation point Z2 included in the gate region G1 (2m−1), and that the moving body includes the gate region G1 (2m−1) and the gate region Gp ( 2m-1) and a hypothesis that it was not observed.

  The estimation unit 13 obtains a new state vector based on the state vector representing the wake Tr1 and the information on the observation point Z2, thereby indicating that a moving body is observed at the observation point Z1 and the observation point Z2. Estimate a new first wake. A curve Tr2 indicated by a one-dot chain line in FIG. 2 is an example of a wake estimated by the estimation unit 13 using information on the observation point Z2 as a new first wake. Further, the estimation unit 13 uses the state vector representing the wake Tr1 shown in FIG. 2 as a second wake indicated by a hypothesis that the moving body is not observed in the gate region G1 (2m−1). A track obtained by extending the track Tr1 shown in FIG. Further, the estimation unit 13 uses the wake Trp as a second wake indicated by a hypothesis that the moving body is not observed in the gate region Gp (2m−1) based on the state vector representing the wake Trp shown in FIG. Estimate the wake obtained by extending.

  Then, when the partial area dA_2m is photographed, the prediction unit 11 moves the moving body according to each wake based on the wakes indicated by the hypotheses generated in the process until the photographing of the partial area dA_2m-1. A gate region including the predicted position is obtained. That is, the prediction unit 11 predicts the gate area at the time when the partial area dA_2m is captured based on the tracks Tr1, Tr2 and the track Trp. An ellipse Gp (2m) indicated by a broken line in FIG. 2 indicates a gate region predicted by the prediction unit 11 based on the second wake obtained by extending the past wake Trp at the time when the partial region dA_2m is photographed.

  A point Z3 included in the partial area dA_2m illustrated in FIG. 2 is an example of an observation point detected by the image processing unit GP illustrated in FIG. 1 from an image obtained by photographing the partial area dA_2m. In the example of FIG. 2, the observation point Z3 detected in the image dA_2m is the gate region Gp predicted by the prediction unit 11 based on the second wake obtained by extending the past wake Trp at the time when the partial region dA_2m was photographed. (2m). Therefore, the information of the observation point Z3 is passed to the estimation unit 13 as the observation point extracted from the gate region Gp (2m) by the extraction unit 12.

  On the other hand, the gate region G1 (2m) and the gate region G2 (2m) predicted based on the wakes Tr1 and Tr2 shown in FIG. 2 are substantially equal to the gate region G1 (2m−1), and the gate region G1 (2m) and G2 (2m) do not have a portion overlapping with the partial region dA_2m. For this reason, in FIG. 2, illustration of the gate region G1 (2m) and the gate region G2 (2m) is omitted. Further, the extraction unit 12 does not pass the observation point Z3 to the estimation unit 13 as an observation point in the gate region G1 (2m) or the gate region G2 (2m).

  In this case, the estimation unit 13 assumes that a moving body is observed at the observation point Z3 included in the gate region Gp (2m), and that the moving body is a gate region Gp (2m), G1 (2m), G2 (2m ) To generate a hypothesis that it was not observed.

  Then, the estimation unit 13 obtains a new state vector based on the state vector representing the wake Trp and the information of the observation point Z3, so that the moving body is not observed at the observation points Z1 and Z2, and is observed at the observation point Z3. A new first wake indicated by a hypothesis that it has been made is estimated. A curve Tr3 indicated by a two-dot chain line in FIG. 2 is an example of a wake estimated by the estimation unit 13 using information on the observation point Z3 as a new first wake. Further, the estimation unit 13 shows the second wake shown in FIG. 2 as the second wake indicated by the hypothesis that the moving body was not observed in the gate region Gp (2m) based on the state vector representing the wake Trp shown in FIG. Estimate the wake which extended the wake Trp. Further, the estimation unit 13 shows the second wake shown in FIG. 2 as the second wake indicated by the hypothesis that the moving body was not observed in the gate region G1 (2m) based on the state vector representing the wake Tr1 shown in FIG. The wake obtained by extending the wake Tr1 is estimated. Further, the estimation unit 13 extends the wake Tr2 as the second wake indicated by the hypothesis that the moving body is not observed in the gate region G2 (2m) based on the state vector representing the wake Tr2 shown in FIG. To estimate the wake obtained.

  As described above, the tracking device 10 shown in FIG. 1 uses the observation points extracted by the extraction unit 12 using the gate region predicted by the prediction unit 11 every time each partial region dA is photographed. Thus, the estimation unit 13 estimates the track of the moving body, and uses the estimated track for the prediction of the subsequent gate region. Thereby, the estimation unit 13 includes a plurality of tracks including a wake indicated by a hypothesis that the moving body is observed at the time when each of the moving objects is photographed in the plurality of partial regions dA as in the wake Tr2 illustrated in FIG. The wake can be estimated.

  The tracking unit 14 illustrated in FIG. 1 is higher than the other wakes among a plurality of wakes including the first wake confronted by the estimation unit 13 in the process in which each partial area dA is imaged by the imaging device EQ. A wake with likelihood is selected as a wake that is likely to be a moving object. In the example of FIG. 2, the tracking unit 14 is the most likely among the past track Trp of the moving body and the track Tr1, the track Tr2, and the track Tr3 newly generated by the estimation unit 13 as the first track. Tracking processing is performed by selecting at least one wake including a high wake.

  Here, since the observation points Z1 and Z2 shown in FIG. 2 are in the center of the gate regions Gp (2) and G1 (2m−1) used for extraction, the observation points Z1 and Z2 Z2 is highly likely to indicate a moving object. On the other hand, since the observation point Z3 shown in FIG. 2 is in the peripheral part of the gate region Gp (2m), the possibility that the observation point Z3 indicates a mobile object is that the observation points Z1 and Z2 indicate a mobile object. Less than possible. Therefore, when the estimator 13 estimates wakes indicating the wakes Trp, Tr1, Tr2, and Tr3 shown in FIG. 2, the tracking unit 14 selects the wake Tr2 as the most likely wake.

  The tracking unit 14 includes a plurality of hypotheses including a hypothesis indicating the first wake generated by the estimation unit 13 in a process of capturing each partial region dA by the imaging device EQ in a tree structure indicating the relationship between the hypotheses. The hypothesis tree is generated by positioning. The tracking unit 14 evaluates the likelihood of each hypothesis included in the hypothesis tree as the likelihood of the wake indicated by each hypothesis based on the hierarchical structure indicated by the generated hypothesis tree. Note that a method of calculating the likelihood of each hypothesis in the tracking unit 14 will be described later with reference to FIGS.

  Further, FIG. 1 illustrates a process in which the tracking unit 14 deletes other hypotheses while leaving, for example, a predetermined number of hypotheses including the hypothesis with the highest likelihood based on the likelihood obtained for each hypothesis. The hypothesis is selected by performing each time one frame image corresponding to the observed area Ar is acquired.

  FIG. 3 shows an example of a hypothetical tree. The hypothesis tree shown in FIG. 3 is a transition from the hypothesis indicating the past wake Trp of the moving body shown in FIG. 2 by the estimation unit 13 in the process in which the observation area Ar shown in FIG. Includes hypotheses generated as possible hypotheses. In the hypothesis tree shown in FIG. 3, of the hypotheses that can be transitioned from the hypothesis indicating the past track of the moving object, the hypothesis that the observation point is noise or the observation point is a new target to be tracked. The hypothesis that it is a mobile object is not shown.

  In the hypothesis tree shown in FIG. 3, each hypothesis generated by the estimation unit 13 shown in FIG. 1 is indicated by a circle. Further, in the codes indicating the respective hypotheses, the numbers following the code “h” indicate the respective hypotheses generated by the estimating unit 13 illustrated in FIG. 1 in the process of taking a picture of each partial area dA included in the observation area Ar. Indicates the number to distinguish. In FIG. 3, generated when each partial area dA is photographed on the dotted lines indicating the times t_1, t_2, t_2m-1, t_2m when the partial areas dA_1, dA_2, dA_2m-1, and dA_2m are photographed. The hypotheses are shown side by side. Further, the double circle hp shown in FIG. 3 indicates a hypothesis indicating the past wake Trp obtained by the tracking process up to the previous frame.

  Further, in the hypothesis tree shown in FIG. 3, the hypotheses indicated by white circles h2, h4, and h9 are hypotheses that a moving object is observed at the extracted observation points, and information on the observation points is used. The first wake obtained by updating the state vector of the wake indicated by the hypothesis corresponding to the parent is shown. For example, in FIG. 3, the hypothesis h2 indicated by a white circle on the dotted line indicating the time t_2 is a hypothesis corresponding to the child of the hypothesis h1 indicating the wake Trp shown in FIG. 2, and the moving object moving according to the wake Trp is observed. This is a hypothesis that it is observed at the point Z1. The first wake indicated by this hypothesis h2 is the wake Tr1 represented by the state vector obtained by updating the state vector representing the wake Trp using the information of the observation point Z1. Similarly, in FIG. 3, a hypothesis h4 indicated by a white circle on the dotted line indicating time t_2m-1 is a hypothesis corresponding to a child of hypothesis h2 indicating the wake Tr1, and the moving body moving according to the wake Tr1 is the observation point Z2. It is a hypothesis that it was observed in. The first wake indicated by this hypothesis h4 is the wake Tr2 represented by the state vector obtained by updating the state vector representing the wake Tr1 using the information of the observation point Z2. In FIG. 3, hypothesis h9 indicated by a white circle on the dotted line indicating time t_2m is a hypothesis corresponding to a child of hypothesis h6 indicating wake Trp, and a moving body moving according to wake Trp is observed at observation point Z3. This is a hypothesis. The first wake indicated by this hypothesis h9 is a wake Tr3 represented by a state vector obtained by updating the state vector representing the wake Trp using information on the observation point Z3.

  On the other hand, in the hypothesis tree shown in FIG. 3, the hypothesis indicated by the shaded circle is a hypothesis that no moving object is observed in the captured partial area dA, and the wake indicated by the hypothesis corresponding to the parent is The 2nd wake which is the wake which inherited the state vector to represent is shown. For example, the hypotheses h1, h3, h6, and h10 indicated by the shaded circles in FIG. 3 are all connected in series to the hypothesis hp that indicates the past wake Trp obtained by the tracking process up to the previous frame. It is a hypothesis that corresponds to the offspring. That is, these hypotheses h1, h3, h6, and h10 are hypotheses that a moving body that moves according to the wake Trp shown in FIG. 2 is not observed in the partial areas dA_1, dA_2, dA_2m-1, and dA_2m. The wakes indicated by these hypotheses h1, h3, h6, and h10 are all wakes represented by a state vector that inherits the speed and acceleration contained in the state vector representing the past wake Trp. Further, in FIG. 3, hypotheses h5 and h8 indicated by circles shaded on dotted lines indicating times t_2m-1 and t_2m are descendants connected in series to hypothesis h2 indicating the wake Tr1 shown in FIG. This is a hypothesis. That is, these hypotheses h5 and h8 are hypotheses that a moving body that moves according to the wake Tr1 indicated by the hypothesis h2 corresponding to the parent is not observed in the partial areas dA_2m-1 and dA_2m. The wake indicated by these hypotheses h5 and h8 is a wake that has inherited the speed and acceleration of the state vector representing the wake Tr1. Similarly, in FIG. 3, a hypothesis h7 indicated by a circle shaded on a dotted line indicating time t_2m is a hypothesis corresponding to a child of hypothesis h4 indicating the wake Tr2 shown in FIG. That is, the hypothesis h7 is a hypothesis that a moving body that moves according to the wake Tr2 indicated by the hypothesis h4 corresponding to the parent is not observed in the partial region dA_2m. The wake indicated by hypothesis h7 is a wake that inherits the speed and acceleration of the state vector representing the wake Tr2. Note that the examples illustrated in FIGS. 2 and 3 illustrate a case where the observation point is not detected from the partial region dA_1 by the image processing unit GP illustrated in FIG.

  For example, the tracking unit 14 illustrated in FIG. 1 compares the likelihoods obtained for the hypotheses h7, h8, h9, and h10 indicated on the dotted line indicating the time t_2m in the hypothesis tree illustrated in FIG. The track of a moving object is tracked by selecting a hypothesis that has a higher likelihood than the hypothesis. For example, when the likelihood of hypothesis h7 is the highest, the tracking unit 14 outputs the wake Tr2 indicated by the hypothesis h7 as a likely wake of the moving object obtained by observing the observation region Ar over the past k times. To do.

  In addition, the tracking unit 14 selects a hypothesis by performing a “pruning” process in which a branch including a hypothesis selected as a hypothesis indicating a likely wake is left in the hypothesis tree and other branches are deleted. Also good. For example, the tracking unit 14 estimates the likelihood of the hypothesis included as an element for each of the branch B_1 including the hypotheses h2 and each hypothesis branching from the hypothesis h2, and the branch B_2 including each node branching from the hypothesis h3 and the hypothesis h3. Find the sum of. For example, when the sum of the likelihoods obtained for the branch B_1 is larger than the sum of the likelihoods obtained for the branch B_2, each hypothesis included in the branch B_1 is left and each hypothesis included in the other branch B_2 is deleted. By doing so, the “pruning” process is performed.

  In addition, the tracking unit 14 notifies the prediction unit 11 of the hypothesis selected in the “pruning” process, and in the subsequent tracking process, the prediction unit 11 predicts the gate region using the track indicated by the selected hypothesis. It is desirable to do it. Thereby, tracking of the moving body can be performed in consideration of information included in an image captured after that, together with an image of one frame obtained by imaging the observation region Ar for the kth time.

  As described above, the estimation unit 13 included in the tracking device 10 illustrated in FIG. 1 is configured such that when an observation point is extracted from the gate region obtained by the prediction unit 11, a mobile object is extracted from the extracted observation point. Generates a hypothesis that is observed. In addition, the generation unit 13 estimates a wake having a new state vector obtained by updating the state vector of the past wake using the observation point information as the first wake indicated by the generated hypothesis. Then, based on the new first wake estimated by the estimation unit 13, the prediction unit 11 obtains a gate region to be used when extracting an observation point from the partial region dA to be photographed thereafter.

  For example, based on the wake Tr1 obtained by updating the past wake Trp using the information on the observation point Z1 extracted from the partial area dA_2 illustrated in FIG. 2, the prediction unit 11 uses the partial area dA_2m−1. The gate region G1 (2m-1) used when extracting the observation points from is obtained. Then, the estimation unit 13 further updates the wake Tr1 using the information on the observation point Z2 extracted from the gate region G1 (2m-1), so that a new first wake that becomes one of the conceivable wakes of the moving object is obtained. A wake Tr2 is estimated as a wake. That is, the estimation unit 13 shown in FIG. 1 uses the wake Tr2 shown in FIG. 2 as a wake Tr2 in which the information of the observation points Z1 and Z2 extracted from the two partial areas dA_2 and dA_2m−1 is reflected. Can be estimated.

  Here, the observation point Z2 shown in FIG. 2 is included in the gate region G1 (2m−1) obtained based on the wake Tr1, but the past point obtained by the tracking process up to the previous frame is included. It is not included in the gate region Gp (2m−1) obtained based on the wake Trp. That is, the wake Tr2 obtained by updating the wake Tr1 using the information on the observation point Z2 is not generated by the tracking processing by the conventional tracking device, but is generated for the first time by the estimation unit 13 of the tracking device 10 shown in FIG. Is done. 2 is compared with the other tracks shown in FIG. 2 when the hypothesis that the mobile is observed at both the observation points Z1 and Z2 is true. It becomes the wake closest to the true wake.

  That is, the tracking unit 14 shown in FIG. 1 includes hypotheses indicating possible wakes of the moving object, including the hypothesis that the moving object is observed at the observation points extracted from the plurality of partial areas dA in the observation area Ar. You can receive a set of hypotheses that contain without omissions.

  The tracking unit 14 shown in FIG. 1 selects a plausible hypothesis from a set of hypotheses that completely include a hypothesis indicating a possible wake of the mobile object, and the wake indicated by the selected hypothesis is Select as a likely wake. Thereby, the tracking device 10 shown in FIG. 1 can move the moving body during the shooting of the plurality of partial areas dA in response to a change in the speed of the moving body that occurs within the time required for shooting the observation area Ar. Even when the wake of the vehicle changes, the track of the moving body can be tracked with higher accuracy than in the past.

  FIG. 4 shows the operation of the tracking device 10 shown in FIG. The process of step S301 to step S307 shown in FIG. 4 shows the operation of the tracking device 10 shown in FIG. 1 and an embodiment of the tracking method and the tracking program. For example, the process illustrated in FIG. 4 is realized by a processor installed in the tracking device 10 executing a tracking program. Note that the processing illustrated in FIG. 4 may be executed by hardware mounted on the tracking device 10.

  The tracking device 10 executes the processing in steps S301 to S307 shown in FIG. 4 every time the observation region Ar is imaged by the imaging device EQ, for example.

  In step S301, the estimation unit 13 observes extracted by the image processing unit GP from images obtained by sequentially capturing each of the partial regions dA included in the observation region Ar illustrated in FIG. 1 by the infrared sensor SC. Receive point information.

  In step S302, the prediction unit 11 obtains a gate region including the predicted position of the moving object at the time when the partial region was captured in step S301, based on one of the wakes generated so far.

  In step S303, the extraction unit 12 determines whether the observation point received in step S301 is included in the gate region obtained in step S302.

  When the observation point is in the gate region (Yes in step S303 (YES)), the extraction unit 12 passes the information indicating the observation point received in the process of step S301 to the estimation unit 13, and based on the received information. The estimation unit 13 executes the process of step S304.

  In step S304, the estimation unit 13 estimates a new first track using the track and observation point information used for the prediction of the gate region in the process of step S302. In the process of step S304, the estimation unit 13 indicates that the moving object is not observed in the gate region used for extraction of the observation point together with the hypothesis indicating the first wake that the moving object is observed at the extracted observation point. Generate a hypothesis to Further, the estimation unit 13 performs estimation of the second wake indicated by a hypothesis that the moving object is not observed in the gate region used for extraction of the observation point, together with the estimation of the new first wake.

  After the end of the process in step S304 or in the case of a negative determination (NO) in step S303, the process proceeds to the process in step S305.

  In step S305, the prediction unit 11 determines whether or not the process of predicting the gate area has been executed for all the tracks estimated so far. For example, the prediction unit 11 determines whether or not the gate area has been obtained for all the wakes estimated up to the capturing of the partial area dA performed before the partial area dA captured in the process of step S301.

  When there is a hypothesis that has not yet obtained a gate region (No determination in step S305 (NO)), the prediction unit 11 returns to the process of step S302 and performs a process of obtaining a gate region for one of the unprocessed hypotheses. Run.

  When the prediction of the gate region for all hypotheses and generation of the hypothesis using the predicted gate region are completed by repeating the processing of step S302 to step S305 (Yes determination in step S305 (YES)) Shifts to the process of step S306.

  In step S306, the estimation unit 13 determines whether or not imaging of all partial areas dA included in the observation area Ar has been completed.

  If there is a partial area dA that has not been shot yet (No determination in step S306 (NO)), the process returns to the process of step S301, and the extraction unit 12 responds to the shooting of the new partial area dA by the shooting apparatus EQ. Thus, the information of the observation point extracted from the photographed partial area dA is received.

  When the processes for all the partial areas dA included in the observation area Ar are completed by repeating the processes of Step S301 to Step S306 (Yes in Step S306 (YES)), the tracking unit 14 performs the process of Step S307. Execute the process.

  In step S307, the tracking unit 14 selects a hypothesis that is more likely than the other hypotheses from among a plurality of hypotheses generated by the estimation unit 13 in the process of repeating the processes of steps S301 to S306, and moves. Select a wake that is likely to be in your body.

  As described with reference to FIGS. 2 and 3, the set of hypotheses generated by repeating the processes in steps S301 to S306 described above is a track in which a moving body is observed in a plurality of partial areas dA. Including the hypothesis to show, the hypothesis showing the possible wake is included without omission.

  Therefore, the tracking unit 14 can track the track of a moving object with higher accuracy than before by selecting a plausible hypothesis from a set of hypotheses generated by repeating the processes of steps S301 to S306. It is.

  Note that the prediction unit 11 illustrated in FIG. 1 may execute the process of step S302 illustrated in FIG. 4 in parallel on all the tracks estimated so far. Further, the extraction unit 12 and the estimation unit 13 illustrated in FIG. 1 may execute the processes of step S303 and step S304 using each of the gate regions obtained in parallel by the prediction unit 11.

  In addition, when a plurality of moving objects exist in the observation area Ar shown in FIG. 1, the tracking device 10 generates a cluster that summarizes hypotheses indicating the track of the moving object, for example, corresponding to each of the moving objects. Then, it is desirable to execute the tracking process described in FIG. 4 for each cluster.

  FIG. 5 shows another embodiment of the tracking device 10. 5 that are equivalent to the components shown in FIG. 1 are denoted by the same reference numerals, and the description thereof may be omitted.

  The tracking device 10 illustrated in FIG. 5 includes an output unit 15 in addition to the prediction unit 11, the extraction unit 12, the estimation unit 13, and the tracking unit 14 illustrated in FIG. The tracking unit 14 illustrated in FIG. 5 includes an evaluation unit 141 and a selection unit 142.

  For example, the evaluation unit 141 evaluates the likelihood of each hypothesis generated by the estimation unit 13 in the process of imaging each partial area dA included in the observation area Ar by the imaging apparatus EQ.

  The evaluation unit 141 uses the partial area dA from which the observation point used to generate the first wake is extracted and the gate area used to extract the observation point as the likelihood of each hypothesis indicating the first wake. The larger the area, the larger the value. Further, the evaluation unit 141 sets the likelihood of each hypothesis indicating the second wake as a larger value as the area of the portion where the captured partial region dA overlaps the gate region used for extraction of the observation point is smaller. .

  For example, the evaluation unit 141 uses the likelihood obtained for the hypothesis indicating the wake used for generating the first wake and the generation of the first wake to evaluate the likelihood of each hypothesis indicating the first wake. And the probability that a moving object is observed in the partial area from which the observed observation points are extracted. For example, the evaluation unit 141 evaluates the likelihood of the hypothesis h2 shown in the hypothesis tree of FIG. 3 by observing the likelihood of the hypothesis h1 corresponding to the parent of the hypothesis h2 and the partial area dA_2 shown in FIG. Probability of being used.

  Further, the evaluation unit 141 evaluates the likelihood of the hypothesis indicating each of the second wakes, and moves the likelihood obtained for the hypothesis indicating the wake used to generate the second wake and the captured partial area. The probability that the body is not observed is used. For example, the evaluation unit 141 evaluates the likelihood of the hypothesis h3 shown in the hypothesis tree of FIG. 3, and the mobile object is observed in the likelihood of the hypothesis h1 corresponding to the parent of the hypothesis h3 and the partial area dA_2 shown in FIG. Probability of not being used. Here, the probability that the moving object is not observed in the partial area dA_2 shown in FIG. 2 can be obtained as a value obtained by subtracting the probability that the moving object is observed in the partial area dA_2 shown in FIG.

  By using the probability that the moving object is observed in the partial area dA and the probability that the moving object is not observed, the evaluation unit 141 illustrated in FIG. 5 captures the individual partial areas dA before the imaging of all the partial areas dA is completed. The likelihood of each hypothesis generated at the time can be evaluated.

  Note that the evaluation unit 141 evaluates the likelihood of each hypothesis, including the method used to calculate the probability that a moving object will be observed in the partial region from which the observation points used to generate the first wake are extracted. Will be described with reference to FIGS.

  In addition, the selection unit 142 illustrated in FIG. 5 is obtained by the evaluation unit 141 evaluating the likelihood of the hypotheses generated so far, for example, every time imaging of each partial area dA by the imaging apparatus EQ is completed. Compare the likelihood of each hypothesis. Then, the selection unit 142 selects at least one hypothesis having a higher likelihood than other hypotheses based on the comparison result, and obtains the wake indicated by the selected at least one hypothesis from the tracking results so far. As the past wake, it is passed to the prediction unit 11 and the display device DSP.

  The selection unit 14 compares, for example, the likelihoods obtained by the evaluation unit 141 for each of the hypotheses h7, h8, h9, and h10 corresponding to the leaves of the hypothesis tree illustrated in FIG. 3, and includes at least a hypothesis with the highest likelihood. A wake indicated by one hypothesis may be selected. The selection unit 14 may select a wake by performing the “pruning” process described with reference to FIG. 3.

  In addition, the output unit 15 illustrated in FIG. 5 displays information indicating the likelihood obtained for each hypothesis by the evaluation unit 141 in the process in which the plurality of partial areas dA illustrated in FIG. And a hypothesis whose likelihood is indicated by the received information to be greater than or equal to the second threshold is detected. Further, each time the hypothesis whose likelihood is shown to be greater than or equal to the second threshold is detected, the output unit 15 passes information representing the wake indicated by the detected hypothesis to, for example, the display device DSP, and the detected hypothesis Display the wake shown.

  For example, the second threshold value used in the output unit 15 is set to a value similar to the threshold value used when determining whether or not there is a high possibility of showing the track of the moving object based on the likelihood obtained for the hypothesis. It is desirable to do.

  In the tracking device 10 shown in FIG. 5, the evaluation unit 141 evaluates the likelihood of each hypothesis generated by the estimation unit 13 at the time of photographing each partial region dA, and the output unit 15 uses the evaluation unit 141 to A wake indicated by a hypothesis that has a likelihood equal to or greater than the second threshold is output via the display device DSP.

  Thus, before the imaging of the 2m partial areas dA is completed, the wakes estimated to have higher likelihood than the other wakes among the wakes estimated at the time of imaging each of the partial areas dA are detected. The detected wake can be displayed on the display device DSP as a likely wake of the moving object. That is, the tracking device 10 shown in FIG. 5 displays the likely track of the moving body on the display device DSP without waiting for the completion of the imaging of all the partial regions dA included in the observation region Ar. The latest information can be presented to the operator at an early stage.

  By the way, the probability that the moving body is observed in each partial area dA is the probability that the moving body exists in the portion where the gate area and the partial area dA obtained when the partial area dA is photographed, Is obtained as a product of the probability Pd observed in the observation region Ar. The probability Pd that the moving body is observed in the observation area Ar is a constant set to a value close to the numerical value 1 of about 0.9 or higher, for example. On the other hand, the probability that the moving body exists in the portion where the gate region and the partial region dA overlap is obtained by integrating a function indicating the probability density of the moving body in the portion where the gate region and the partial region dA overlap. Desired.

  Here, the gate area at the time when each partial area dA is photographed is obtained for each of the different wakes estimated by the estimation unit 13 so far. For example, when L (L is a positive integer) new wakes are estimated in the process until the j-th partial area dA_j is photographed in the observation area Ar shown in FIG. The prediction unit 11 shown determines the gate area based on the past track Trp and L new tracks.

  In the following description, the past track Trp and the L new tracks are represented by a code “Tr” indicating the track, a code “p” indicating the past track, or a number for distinguishing each new track. This is indicated by a reference symbol Trc in combination with a reference symbol “c”. Further, based on the wake Trc, the gate area obtained for the time t_j at which the partial area dA_j was photographed is a code “G” that is a combination of the code “G” indicating the gate area and the code “c” for distinguishing each wake. Gc ″ is used to denote a gate region Gc (j). A portion where the gate region Gc (j) and the partial region dA_j overlap is referred to as an overlapping portion Sc_j. Also, when M observation points are extracted by the extraction unit 12 shown in FIG. 1 or 5 in the process of taking a plurality of partial areas dA, each observation point has a symbol “ Z ″ and a number i (i is a positive integer less than or equal to an integer M) that distinguishes individual observation points are used as observation points Zi.

  The evaluation unit 141 uses the probability Pc_j that the moving body moving according to the wake Trc exists in the overlapping portion Sc_j where the partial region dA_j and the gate region Gc (j) overlap at the time tj when the partial region dA_j is captured. For example, it is calculated by the equation (1).

なお、式(１)において、関数Ｎｃ＿ｊ(ｘ，ｙ)は、航跡Ｔｒｃに基づいて時刻ｔｊについて求められたゲート領域Ｇｃ(ｊ)における移動体の存在の確率密度であり、航跡Ｔｒｃに従う移動体の時刻ｔｊにおける予測位置を期待値とする２次元の正規分布を示す。

In equation (1), the function Nc_j (x, y) is the probability density of the existence of the moving object in the gate region Gc (j) obtained for the time tj based on the wake Trc, and the moving object according to the wake Trc. 2D shows a two-dimensional normal distribution with the predicted position at time tj as an expected value.

  The evaluation unit 141 shown in FIG. 5 uses the probability Pc_j calculated by the expression (1) and the hierarchical structure between hypotheses shown by the hypothesis tree shown in FIG. 3, and the likelihood of the wake shown by each hypothesis. Assess degree. For example, when the hypothesis hb to be evaluated is a hypothesis indicating the first wake indicated by a white circle in the hypothesis tree illustrated in FIG. 3, the evaluation unit 141 estimates the likelihood L (hb) of the hypothesis. Is evaluated using equation (2).

式(２)において、数値Ｌ(ｈｆ)は、評価の対象となる仮説ｈｂの親に当たる仮説ｈｆについて求められた尤度を示す。ここで、評価の対象となる仮説ｈｂの親に当たる仮説とは、仮説ｈｂが生成される際のゲート領域Ｇｃ(ｊ)を求める際に用いられた航跡Ｔｒｃを示す仮説である。例えば、図３に示した仮説ｈ２の親の仮説である仮説ｈ１は、図２に示したゲート領域Ｇｐ(２)の算出に用いられた、過去の航跡Ｔｒｐを延長した航跡を示す仮説である。また、式(２)において、数値Ｎｃ＿ｊ(Ｘｚｉ，Ｙｚｉ)は、観測点Ｚｉにおける移動体の存在の確率密度を示す。また、式(２)において、式(１)で求められた確率Ｐｃ＿ｊと確率Ｐｄとの積は、仮説ｈｆで示される航跡Ｔｒｃで移動する移動体が部分領域ｄＡ＿ｊで観測される確率を示す。即ち、式(２)にて、仮説ｈｂの尤度Ｌ(ｈｂ)は、親の仮説ｈｆの尤度Ｌ(ｈｆ)と、仮説ｈｆで示される航跡Ｔｒｃで移動する移動体が部分領域ｄＡ＿ｊで観測される確率と、観測点Ｚｉにおける移動体の存在の確率密度Ｎｃ＿ｊ（Ｘｚｉ，Ｙｚｉ）との積で示される。

In Expression (2), a numerical value L (hf) indicates the likelihood obtained for the hypothesis hf corresponding to the parent of the hypothesis hb to be evaluated. Here, the hypothesis corresponding to the parent of the hypothesis hb to be evaluated is a hypothesis indicating the wake Trc used when obtaining the gate region Gc (j) when the hypothesis hb is generated. For example, the hypothesis h1, which is the hypothesis of the parent of the hypothesis h2 shown in FIG. 3, is a hypothesis indicating a wake obtained by extending the past wake Trp used for the calculation of the gate region Gp (2) shown in FIG. . In Equation (2), the numerical value Nc_j (Xzi, Yzi) indicates the probability density of the existence of the moving object at the observation point Zi. In Equation (2), the product of the probability Pc_j and the probability Pd obtained in Equation (1) indicates the probability that a moving body moving on the wake Trc indicated by the hypothesis hf is observed in the partial region dA_j. That is, in Equation (2), the likelihood L (hb) of the hypothesis hb is the likelihood L (hf) of the parent hypothesis hf and the moving body moving in the wake Trc indicated by the hypothesis hf is the partial area dA_j. It is represented by the product of the observed probability and the probability density Nc_j (Xzi, Yzi) of the existence of the moving object at the observation point Zi.

  Further, when the hypothesis hb to be evaluated is a hypothesis indicating the second wake indicated by a circle shaded in the hypothesis tree illustrated in FIG. 3, the evaluation unit 141 calculates the likelihood L of the hypothesis. (hb) is evaluated using equation (3).

式(３)において、数値Ｌ(ｈｆ)は、式(２)と同じく、評価の対象となる仮説ｈｂの親に当たる仮説ｈｆについて求められた尤度を示す。また、式(３)において、式(１)で求められた確率Ｐｃ＿ｊと確率Ｐｄとの積を数値１から減じた値は、仮説ｈｆで示される航跡Ｔｒｃで移動する移動体が部分領域ｄＡ＿ｊで観測されない確率を示す。即ち、式(３)にて、仮説ｈｂの尤度Ｌ(ｈｂ)は、親の仮説ｈｆの尤度Ｌ(ｈｆ)と、仮説ｈｆで示される航跡Ｔｒｃで移動する移動体が部分領域ｄＡ＿ｊで観測されない確率との積で示される。

In Equation (3), the numerical value L (hf) indicates the likelihood obtained for the hypothesis hf corresponding to the parent of the hypothesis hb to be evaluated, as in Equation (2). Further, in the equation (3), the value obtained by subtracting the product of the probability Pc_j and the probability Pd obtained in the equation (1) from the numerical value 1 indicates that the moving body moving on the wake Trc indicated by the hypothesis hf is the partial region dA_j. Indicates the probability of not being observed. That is, in Equation (3), the likelihood L (hb) of the hypothesis hb is the likelihood L (hf) of the parent hypothesis hf and the moving body moving in the wake Trc indicated by the hypothesis hf is the partial area dA_j. It is shown as the product of the probability of not being observed.

  By using the equation (2), the gate region Gc, which is a target range for calculating the probability that a moving object exists using the equation (1), as the likelihood L (hb) of the hypothesis hb indicating the first wake. A larger value can be obtained as the overlapping portion Sc_j between (j) and the partial region dA_j is larger. On the other hand, by using Equation (3), a gate that is a target range for calculating the probability that a moving object exists using Equation (1) as the likelihood L (hb) of hypothesis hb indicating the second wake. A larger value can be obtained as the portion where the region Gc (j) and the partial region dA_j overlap is smaller.

  In the following, the probability that the moving object is observed in each of the partial areas dA_1, dA_2, dA_2m-1, and dA_2m in the case where the moving object moves according to the tracks Trp and Tr1 shown in FIG. An example is described in which the likelihood of a hypothesis is evaluated using.

  FIG. 6 shows an example of calculating the likelihood of each hypothesis shown in FIG. In the table shown in FIG. 6, each symbol included in the column indicated by “hypothesis” indicates hypothesis h <b> 1 to h <b> 10 included in the hypothesis tree of FIG. 3. Further, in the table shown in FIG. 6, the formulas L (h1), L (h2), L (h3), L (h4), L (h5), L included in the column indicated by “likelihood of hypothesis”. Each of (h6), L (h7), L (h8), L (h9), and L (h10) indicates the likelihood of each of hypotheses h1 to h10.

  7 to 9 show examples of ranges in which the probability density of the existence of the moving object is integrated in Equation (1). 7-9, illustration of partial areas dA other than partial areas dA_1, dA_2, dA_2m-1, and dA_2m out of 2m partial areas dA included in observation area Ar shown in FIG. 1 is omitted. ing. 7 to 9, the x-axis indicates the horizontal direction and the y-axis indicates the vertical direction. In addition, among the elements illustrated in FIGS. 7 to 9, elements equivalent to those illustrated in FIG. 2 are denoted by the same reference numerals, and description thereof may be omitted.

  FIG. 7A shows a range in which the probability density of the existence of a moving object is integrated in Equation (1) when calculating the likelihood of the hypothesis h1 shown in FIG. 3, that is, the moving object moving according to the wake Trp. An overlapping portion Sp_1 between the gate region Gp (1) indicating the existence probability density and the partial region dA_1 is shown. In the example of FIG. 7A, in the gate region Gp (1), the overlapping portion Sp_1 that is a portion included in the partial region dA_1 is shown as a shaded region. In FIG. 7A, a point Qe (Trp, 1) indicated by a white circle indicates a predicted position of the moving body at time t_1 when the partial area dA_1 is imaged, which is calculated based on the wake Trp.

  Since the hypothesis h1 shown in the hypothetical tree of FIG. 3 is a hypothesis indicating the second wake that has inherited the speed and acceleration of the past wake Trp, the evaluation unit 141 shown in FIG. Formula (3) is used for evaluation of (h1).

  Here, since the area of the overlapping portion Sp_1 shown in FIG. 7A is less than a quarter of the area of the gate region Gp (1), the probability Pp_1 that the moving body exists in the overlapping portion Sp_1 is expressed as The value obtained by (1) is smaller than the numerical value 0.25. Therefore, the probability that the moving body moving according to the wake Trp is not observed in the partial region dA_1 is a value larger than the numerical value 0.75. Therefore, the evaluation unit 151 obtains a value larger than the product of the likelihood L (hp) of the hypothesis indicating the past track Trp and the numerical value 0.75 as the likelihood L (h1) of the hypothesis h1.

  For example, when the likelihood L (h1) of the hypothesis h1 is larger than the second threshold described in FIG. 5, the output unit 15 illustrated in FIG. 5 converts the wake Trp indicated by the hypothesis h1 into the partial region dA_1. Is detected as a probable wake at the time of shooting, and is displayed on the display device DSP.

  FIG. 7 (B) shows a range in which the probability density of the existence of a moving object is integrated in equation (1) when calculating the likelihood of hypotheses h2 and h3 shown in FIG. An overlapping portion Sp_2 of the gate region Gp (2) indicating the probability density of the presence of the body and the partial region dA_2 is shown. In the example of FIG. 7B, of the gate region Gp (2) surrounded by the broken line, the overlapping portion Sp_2 that is a portion included in the partial region dA_2 is shown as a shaded region. . In FIG. 7B, a point Qe (Trp, 2) indicated by a white circle indicates the predicted position of the moving body at time t_2 when the partial area dA_2 was imaged, calculated based on the track Trp.

  The hypothesis h2 shown in the hypothesis tree of FIG. 3 is a hypothesis indicating the wake Tr1 that is the first wake obtained by updating the past wake Trp using the information of the observation point Z1, and the hypothesis h3 This is a hypothesis indicating a second wake that inherits the speed and acceleration of the wake Trp. Therefore, the evaluation unit 141 illustrated in FIG. 5 uses Expression (2) for evaluating the likelihood L (h2) of the hypothesis h2, and uses Expression (3) for evaluating the likelihood L (h3) of the hypothesis h3. .

  Since the area of the overlapping portion Sp_2 shown in FIG. 7B is about one-fourth of the area of the gate region Gp (2), the probability Pp_2 that the moving body exists in the overlapping portion Sp_2 is expressed by the equation (1 ) Is about 0.25. Therefore, the likelihood L (h2) of the hypothesis h2 is obtained by multiplying the likelihood L (h1) of the hypothesis h1 corresponding to the parent by 0.25 from the equation (2), and further multiplying the value by the constant Pd by the observation point Z1. The probability density Np_2 (Xz1, Yz1) at is multiplied by a value. Note that in the likelihood L (h2) shown in FIG. 6, the probability density Np_2 (Xz1, Yz1) at the observation point Z1 is the coordinates (Xz1, Yz1) of the observation point Z1 in the normal distribution Np_2 indicating the gate region Gp (2). ) Is the value obtained by substituting. On the other hand, since the probability that a moving body that moves according to the wake Trp is not observed in the partial region dA_2 is about 0.75, the likelihood L (h3) of the hypothesis h3 is set to 0. The value is almost equivalent to the value multiplied by 75.

  For example, if each of the likelihoods L (h2) and L (h3) obtained for the hypotheses h2 and h3 exceeds the second threshold, the output unit 15 shown in FIG. 5 indicates the hypothesis h2. Both the wake Tr1 and the wake Trp indicated by the hypothesis h3 are output as likely wakes.

  In FIG. 8A, when calculating the likelihood of the hypotheses h4 and h5 shown in FIG. An overlapping portion S1_2m-1 between the gate region G1 (2m-1) and the partial region dA_2m-1 for the moving body is shown. In the example of FIG. 8A, among the gate region G1 (2m-1) surrounded by a broken line, the overlapping portion S1_2m-1 that is a portion included in the partial region dA_2m-1 is shaded. Shown as a region. In FIG. 8A, a point Qe (Tr1, 2m-1) indicated by a white circle is a moving body at time t_2m-1 when the partial area dA_2m-1 is imaged, which is calculated based on the wake Tr1. Indicates the predicted position.

  The hypothesis h4 shown in the hypothesis tree of FIG. 3 is a hypothesis indicating the wake Tr2 that is the first wake obtained by updating the wake Tr1 using the information of the observation point Z2, and the hypothesis h5 is the wake Tr1. This is a hypothesis indicating the second wake that has inherited the speed and acceleration. Therefore, the evaluation unit 141 illustrated in FIG. 5 uses Expression (2) to evaluate the likelihood L (h4) of the hypothesis h4 and uses Expression (3) to evaluate the likelihood L (h5) of the hypothesis h5. .

  Since the area of the overlapping portion S1_2m-1 shown in FIG. 8A is almost the entire area of the gate region G1 (2m-1), the probability P1_2m-1 that the moving body exists in the overlapping portion S1_2m-1 is assumed. The value obtained by equation (1) is approximately the numerical value 1. Therefore, the likelihood L (h4) of the hypothesis h4 is obtained from the equation (2) by multiplying the likelihood L (h2) of the hypothesis h2 corresponding to the parent by the constant Pd and the probability density N1_2m−1 (Xz2) at the observation point Z2. , Yz2). In the likelihood L (h4) shown in FIG. 6, the probability density N1_2m-1 (Xz2, Yz2) at the observation point Z2 is the observation point Z2 in the normal distribution N1_2m-1 indicating the gate region G1 (2m-1). This is a value obtained by substituting the coordinates (Xz2, Yz2). In the example of FIG. 8A, the distance between the observation point Z2 and the predicted position Qe (Tr1,2m-1) of the moving object is, for example, compared with the minor axis of the ellipse indicating the gate region G1 (2m-1). short. Therefore, since the probability density at the observation point Z2 is larger than the average probability density in the gate region G1 (2m−1), the likelihood L (h4) of the hypothesis h4 obtained by the evaluation unit 141 is There is a possibility that it becomes larger than the likelihood L (h6) of hypothesis h6 described later. On the other hand, since the probability that the moving body moving according to the wake Trp is not observed in the partial region dA_2 is a value close to the numerical value 0, the likelihood L (h5) of the hypothesis h5 obtained using the equation (3) is the numerical value 0. A close value.

  FIG. 8B shows a range in which the probability density of the existence of a moving object is integrated in equation (1) when calculating the likelihood of the hypothesis h6 shown in FIG. 3, that is, a moving object that moves according to the wake Trp. The overlapping portion Sp_2m-1 of the gate region Gp (2m-1) and the partial region dA_2m-1 is shown. In the example of FIG. 8B, among the gate region Gp (2m−1) surrounded by a broken line, the overlapping portion Sp_2m−1 that is a portion included in the partial region dA_2m−1 is shaded. Shown as a region. In FIG. 8B, a point Qe (Trp, 2m−1) indicated by a white circle is a moving object at time t_2m−1 when the partial area dA_2m−1 is calculated based on the wake Trp. Indicates the predicted position.

  Since the hypothesis h6 shown in the hypothesis tree of FIG. 3 is a hypothesis showing the second wake that has inherited the speed and acceleration of the wake shown in the hypothesis h3, the evaluation unit 141 shown in FIG. Formula (3) is used for evaluation of degree L (h6).

  Here, since the area of the overlapping portion Sp_2m-1 illustrated in FIG. 8B is three-quarters or more of the area of the gate region Gp (2m-1), the moving body is within the overlapping portion Sp_2m-1. The value obtained by the equation (1) as the existence probability Pp_2m−1 is larger than the numerical value 0.75. Therefore, the probability that the moving body that moves according to the wake Trp is not observed in the partial region dA_2m−1 is a value smaller than the numerical value 0.25. For this reason, the evaluation unit 151 obtains a value smaller than the product of the likelihood L (h3) of the hypothesis h3 corresponding to the parent and the numerical value 0.25 as the likelihood L (h6) of the hypothesis h6.

  Of the likelihoods L (h4), L (h5), and L (h6) obtained for the hypotheses h4, h5, and h6, for example, when the likelihood L (h4) is greater than or equal to the second threshold, The unit 15 detects the hypothesis h4 as a probable hypothesis at the time t_2m-1. And the output part 15 outputs wake Tr2 as a probable wake via display device DSP by passing the information which shows wake Tr2 shown by hypothesis h4 to display device DSP.

  FIG. 9A shows a range in which the probability density of the existence of a moving object is integrated in Equation (1) when calculating the likelihoods of hypotheses h7 and h8 shown in FIG. In the example of FIG. 9A, the gate region G1 (2m) surrounded by a broken line and the gate region G2 (2m) surrounded by a dotted line do not have a portion included in the partial region dA_2m. . That is, in the example of FIG. 9A, the gate region G1 (2m) or the gate region G2 (2m) and the overlapping portion S1_2m and S2_2m between the partial region dA_2m do not exist. In FIG. 9A, a point Qe (Tr1, 2m) indicated by a white circle indicates a predicted position at time t_2m at which the partial area dA_2m of the moving body according to the wake Tr1 is captured. In FIG. 9A, illustration of the predicted position at time t_2m when the partial area dA_2m of the moving body according to the wake Tr2 is taken is omitted.

  Since the hypothesis h7 shown in the hypothesis tree of FIG. 3 is a hypothesis indicating the wake shown by the hypothesis h4, that is, the second wake that has inherited the speed and acceleration of the wake Tr2, the evaluation unit 141 shown in FIG. Equation (3) is used to evaluate the likelihood L (h7) of hypothesis h7. Further, since the hypothesis h8 shown in the hypothesis tree of FIG. 3 is a hypothesis indicating the wake shown by the hypothesis h5, that is, the second wake that takes over the speed and acceleration of the wake Tr1 shown in FIG. The evaluation unit 141 shown in FIG. 4 uses Expression (3) for evaluating the likelihood L (h8) of the hypothesis h8.

  Here, as shown in FIG. 9A, since the area of the overlapping portion S2_2m is equal to the numerical value 0, the value obtained by the equation (1) as the probability P2_2m that the moving body exists in the overlapping portion S2_2m is the numerical value 0. be equivalent to. Therefore, the probability that the moving body that moves according to the wake Tr2 is not observed in the partial region dA_2m is a numerical value 1. For this reason, the evaluation unit 151 obtains the same value as the likelihood L (h4) of the hypothesis h4 corresponding to the parent as the likelihood L (h7) of the hypothesis h7. Similarly, since the area of the overlapping portion S1_2m is equal to the numerical value 0, the value obtained by the equation (1) as the probability P1_2m that the moving body exists in the overlapping portion S1_2m is equal to the numerical value 0. Therefore, the probability that a moving body that moves according to the wake Tr1 is not observed in the partial region dA_2m is a numerical value 1. For this reason, the evaluation unit 151 obtains the same value as the likelihood L (h5) of the hypothesis h5 corresponding to the parent as the likelihood L (h8) of the hypothesis h8.

  In FIG. 9B, when calculating the likelihoods of the hypotheses h9 and h10 shown in FIG. An overlapping portion Sp_2m between the gate region Gp (2m) and the partial region dA_2m for the moving body is shown. In the example of FIG. 9B, of the gate region Gp (2m) surrounded by a broken line, an overlapping portion Sp_2m that is a portion included in the partial region dA_2m is shown as a shaded region. . In FIG. 9B, a point Qe (Trp, 2m) indicated by a white circle indicates a predicted position of the moving body at time t_2m when the partial area dA_2m is imaged, which is calculated based on the wake Trp.

  The hypothesis h9 shown in the hypothesis tree of FIG. 3 is a hypothesis indicating the wake Tr3 that is the first wake obtained by updating the wake Trp using the information of the observation point Z3, and the hypothesis h10 is the wake Trp of the wake Trp. This is a hypothesis indicating the second wake that has inherited the speed and acceleration. Therefore, the evaluation unit 141 illustrated in FIG. 5 uses Expression (2) for evaluating the likelihood L (h9) of the hypothesis h9 and uses Expression (3) for evaluating the likelihood L (h10) of the hypothesis h10. .

  Since the area of the overlapping portion Sp_2m shown in FIG. 9B is smaller than a quarter of the area of the gate region Gp (2m), the probability Pp_2m that the moving body exists in the overlapping portion Sp_2m is expressed by the equation (1). The value obtained by is smaller than the numerical value 0.25. Accordingly, the likelihood L (h9) of the hypothesis h9 is obtained by multiplying the likelihood L (h6) of the hypothesis h6 corresponding to the parent by 0.25 and multiplying the value obtained by multiplying the constant Pd by the observation point Z3 from the equation (2). The probability density Np_2m (Xz3, Yz3) at is multiplied by the value. Note that in the likelihood L (h9) shown in FIG. 6, the probability density Np_2m (Xz3, Yz3) at the observation point Z3 is represented by the coordinates (Xz3, Yz3) of the observation point Z3 in the normal distribution Np_2m indicating the gate region Gp (2m). ) Is the value obtained by substituting. Here, in the example of FIG. 9B, the observation point Z3 is closer to the edge of the gate region Gp (2m) than the predicted position Qe (Trp, 2m) of the moving body. For this reason, the probability density Np_2m (Xz3, Yz3) at the observation point Z3 is smaller than the average probability density in the gate region Gp (2m). On the other hand, since the probability that the moving body that moves according to the wake Trp is not observed in the partial region dA_2m is larger than the numerical value 0.75, the likelihood L (h10) of the hypothesis h10 is changed to the likelihood L (h6) of the hypothesis h6. The value is larger than the value multiplied by the numerical value 0.75.

  The likelihoods L (h7) to L (h10) of the hypotheses h7 to h10 obtained as described in FIGS. 9A and 9B are compared with the second threshold value in the output unit 15 shown in FIG. Is done. For example, when the likelihood L (h7) is obtained as a value larger than the second threshold, the output unit 15 detects the hypothesis h7 as a probable hypothesis at the time t_2m, and the track Tr2 indicated by the hypothesis h7. Is displayed on the display device DSP.

  As described above with reference to FIGS. 6 to 9, the evaluation unit 141 illustrated in FIG. 5 has each likelihood L (h1 at the time when the hypotheses h1 to h10 illustrated in FIG. 3 are generated. ) To L (h10). Then, based on the obtained likelihoods L (h1) to L (h10), the output unit 15 shown in FIG. 5 determines the first of the hypotheses for which the likelihood was obtained at the time when each partial region was captured. A hypothesis for which a likelihood greater than two thresholds is found is detected. Further, the output unit 15 performs a process of displaying information indicating the wake indicated by the detected hypothesis on the display device DSP. On the other hand, for example, after the calculation of the likelihoods L (h1) to L (h10) of the hypotheses h1 to h10 by the evaluation unit 141 is completed, the selection unit 142 is based on the likelihoods L (h1) to L (h10). The most likely hypothesis is selected from the hypotheses h1 to h10.

  The evaluation unit 141 uses the expressions (1) and (2), so that the moving object is observed at the observation point extracted from the overlapping portion as the area of the overlapping portion between the gate region and the captured partial region is large. A large value is obtained as the likelihood of the hypothesis. Further, the evaluation unit 141 uses the equations (1) and (3), and the hypothesis is that the moving body is not observed in the overlapping portion as the area of the overlapping portion between the gate region and the captured partial region is smaller. A large value is obtained as the likelihood of.

  Thereby, for example, the evaluation unit 141 sets the likelihood of the hypothesis that the moving body is observed in the plurality of partial areas dA as in the hypothesis h7 in the observation area Ar as in the hypotheses h8 and h9. A value that can be compared on the same scale as the likelihood of the hypothesis that the moving object is observed only once is obtained. That is, the evaluation unit 141 includes a hypothesis indicating a wake reflecting observation points extracted from the plurality of partial areas dA and a hypothesis indicating a wake reflecting observation points extracted from one of the plurality of partial areas dA. Likelihood that can be determined by comparing each other can be obtained.

  Therefore, the selection unit 142 compares the likelihoods obtained by the evaluation unit 141 with respect to each hypothesis, and thus a hypothesis that is likely from a plurality of hypotheses generated in the process of photographing 2m partial regions dA. Can be selected.

  As described above, the set of hypotheses passed from the estimation unit 13 to the selection unit 142 includes transitions from past wakes, including wakes that have changed in speed in the process of capturing 2m partial areas dA. Contains the hypothesis indicating the possible wake.

  Therefore, the selection unit 142 selects a hypothesis having a higher likelihood than other hypotheses from a set of hypotheses including all hypotheses generated by the estimation unit 13, thereby capturing a plurality of partial regions dA. A probable track can be selected from a plurality of estimated tracks.

  In other words, the tracking device 10 including the evaluation unit 141 that evaluates the likelihood of the hypothesis using the area of the overlapping portion between the gate region and the captured partial region can track the moving object in the process of capturing the observation region Ar. Even when it changes, it is possible to track the track of a moving object with higher accuracy than in the past.

  Note that the method used by the evaluation unit 141 shown in FIG. 5 to calculate the likelihood is not limited to the method shown by the equations (1) to (3), and the gate region and the imaging are used for calculating the likelihood of each hypothesis. Any method may be used as long as it uses the probability that a moving object is present in an overlapping part with the partial area.

  For example, the evaluation unit 141 may use Equation (4) for evaluating the likelihood of a hypothesis that a moving body is not observed in each partial region dA. In the equation (4), the numerical value γ indicating the exponent part of the numerical value (1-Pd) is an enlarged gate region indicating a range in which the moving object may be observed during the shooting of each partial region. The number of partial areas including at least a part of the range obtained based on the past track of the moving object is shown. That is, in Equation (4), the probability that the moving body does not exist in the overlapping portion between the enlarged gate region and each partial region dA, which is obtained during one round of shooting of each partial region, is a numerical value (1-Pd ) To the 1st power of γ. For example, when the enlarged gate region obtained over the entire area of the observation region Ar is captured for the kth time and straddles four partial regions dA, the moving object moves to the overlapping portion of each partial region dA and the gate region. The probability that no exists is indicated by a quarter power of the numerical value (1-Pd).

ここで、数値γが大きい場合に、各部分領域ｄＡと拡大されたゲート領域との重複部分の面積は、数値γが小さい場合に比べて小さくなる。そして、数値γが大きい場合に、重複部分に移動体が存在しない確率を示す値、即ち、数値(１−Ｐｄ)のγ分の１乗の値は、数値γが小さい場合に比べて大きくなる。すなわち、数値(１−Ｐｄ)のγ分の１乗により、重複部分に移動体が存在しない確率を示す式(４)を用いる手法は、ゲート領域と撮影された部分領域との重複部分に移動体が存在する確率を用いる手法の一例である。

Here, when the numerical value γ is large, the area of the overlapping portion between each partial region dA and the enlarged gate region is smaller than when the numerical value γ is small. When the numerical value γ is large, the value indicating the probability that there is no moving object in the overlapping portion, that is, the value of the power of 1 / γ of the numerical value (1-Pd) is larger than when the numerical value γ is small. . That is, the method using the equation (4) indicating the probability that the moving body does not exist in the overlapping portion by the first power of γ of the numerical value (1-Pd) moves to the overlapping portion between the gate region and the captured partial region. It is an example of the technique using the probability that a body exists.

  When Equation (4) is used to evaluate the likelihood of the hypothesis, the likelihood of the hypothesis that the moving object is not observed in all the partial areas dA is expressed by the 1 / γth power of the numerical value (1-Pd). It is the product of a value obtained by synergizing the values over γ times and the likelihood of a hypothesis indicating a past wake. Here, a value obtained by synthesizing the first power of γ of the numerical value (1-Pd) over γ times is equal to the numerical value (1-Pd). Therefore, when the evaluation unit 141 uses Expression (4), the likelihood of the hypothesis that the moving body is not observed in all the partial areas dA is the likelihood of the hypothesis indicating the past track and the numerical value (1-Pd). It is indicated by the product of That is, the evaluation unit 141 can obtain the same value as the hypothesis likelihood that the moving object is not observed in all the partial areas dA as in the case where the hypothesis is generated after the imaging of the observation area Ar is completed. Note that the value indicating the probability that a moving object is not observed in each partial region is not limited to the value shown in Equation (4), but may be a value that is a numerical value (1−Pd) by synergy with each other. For example, the evaluation unit 141 uses a value obtained by raising the numerical value (1-Pd) by the ratio of the portion included in each partial region of the enlarged gate region, and the moving body is not observed in each partial region dA. Hypothesis likelihood may be evaluated.

  The evaluation unit 141 uses the expression (2) to evaluate the likelihood of the hypothesis indicating the first wake generated using the observation points extracted from the captured partial region. The following correction may be performed on the likelihood L (hf) of the hypothesis hf corresponding to the parent of the hypothesis hb.

  The evaluation unit 141, for example, from the hypothesis hf corresponding to the parent of the hypothesis hb to be evaluated to the hypothesis hp indicating the past wake obtained by the tracking processing from the previous frame to the previous frame or the hypothesis indicating another first wake is shown in FIG. Go back to the hypothesis tree shown in. Then, the evaluation unit 141 detects a hypothesis generated without using observation point information in the process of going back up the hypothesis tree. Further, the evaluation unit 141 divides the likelihood L (hf) obtained for the hypothesis hf by the synergy of the “probability that no moving object is observed” multiplied when the likelihood of the detected hypothesis is evaluated. Thus, the likelihood decrease due to the probability that the moving object is not observed is canceled out. Then, the evaluation unit 141 uses the likelihood L (hf) after the correction of the hypothesis hf indicated by the value obtained by the division, and the likelihood L of the hypothesis hb to be evaluated according to the equation (2). (hb) is calculated.

  For example, when evaluating the likelihood L (h2) of the hypothesis h2 shown in the hypothesis tree of FIG. 3, the evaluation unit 141 is the probability that the moving object is not observed for the likelihood L (h1) of the hypothesis h1 corresponding to the parent. Correction for canceling the decrease in likelihood due to. That is, the evaluation unit 141 divides the likelihood L (h1) by the numerical value (1−Pd · Pp_1) multiplied when the likelihood L (h1) is evaluated, whereby “the moving object is a partial region dA_1. And the likelihood decrease due to the “probability that is not observed in the overlapping part of the gate region Gp (1)”. Then, the evaluation unit 141 uses the result obtained by the division for calculating the likelihood L (h2). When correction for canceling the decrease in likelihood due to the probability that the moving object is not observed is applied, the likelihood L (h2) of hypothesis h2 is expressed by Equation (5).

同様に、移動体が観測されない確率による尤度の減少分を打ち消す補正を適用した場合に、仮説ｈ９の尤度Ｌ(ｈ９)は、式(６)で示される。

Similarly, the likelihood L (h9) of the hypothesis h9 is expressed by Expression (6) when correction for canceling the likelihood decrease due to the probability that the moving object is not observed is applied.

以上に説明した補正により、評価部１４１は、先に撮影された１以上の部分領域にて移動体が観測されず、その後に撮影された部分領域にて観測されたとする仮説の尤度として、全ての部分領域の撮影が完了した際に仮説を生成する場合と同等の値を得ることができる。

As a result of the correction described above, the evaluator 141 estimates the likelihood of the hypothesis that the moving object is not observed in one or more partial areas previously captured and is observed in the partial areas captured thereafter. It is possible to obtain a value equivalent to the case where a hypothesis is generated when photographing of all partial areas is completed.

  The tracking device 10 of the present disclosure described above can be realized using a computer device.

  FIG. 10 shows an example of a hardware configuration of the tracking device 10 shown in FIG. 10 that are equivalent to the components shown in FIG. 1 are denoted by the same reference numerals, and the description thereof may be omitted.

  The computer device 20 includes a processor 21, a memory 22, a hard disk device 23, a general-purpose interface 24, a display control unit 25, an input device 26, and an optical drive device 27. The processor 21, the memory 22, the hard disk device 23, the general-purpose interface 24, the display control unit 25, the input device 26, and the optical drive device 27 illustrated in FIG. 10 are connected to each other via a bus. . In addition, the processor 21, the memory 22, the hard disk device 23, and the general-purpose interface 24 are included in the tracking device 10.

  The optical drive device 27 described above can be mounted with a removable disk 28 such as an optical disk, and reads and records information recorded on the mounted removable disk 27.

  The input device 26 illustrated in FIG. 10 is, for example, a keyboard or a mouse. The operator of the tracking device 10 operates the input device 26 to input, for example, an instruction to start tracking processing of the moving body based on an image obtained by imaging by the imaging device EQ. can do.

  The memory 22 illustrated in FIG. 10 stores an application program for the processor 21 to execute the tracking process described with reference to FIGS. 1 to 9 together with the operating system of the computer device 20. An application program for executing the tracking process can be recorded on a removable disk 28 such as an optical disk, for example. Then, an application program for executing the tracking process may be stored in the memory 22 and the hard disk device 23 by mounting the removable disk 28 in the optical drive device 27 and performing a reading process. Further, an application program for executing the tracking process can be downloaded via a communication device (not shown) connected to a network such as the Internet and read into the memory 22 and the hard disk device 23.

  Then, the processor 21 executes the application program stored in the memory 22 to perform the functions of the prediction unit 11, the extraction unit 12, the estimation unit 13, the tracking unit 14, and the output unit 15 illustrated in FIG. Also good. The processor 21 also stores information used when operating as the prediction unit 11, the extraction unit 12, the estimation unit 13, the tracking unit 14, and the output unit 15 and the generated information in the memory 22 or the hard disk device 23. Hold on.

  In addition, when a plurality of moving objects exist in the observation area Ar shown in FIG. 1, the tracking device 10 generates a cluster that summarizes hypotheses indicating the track of the moving object, for example, corresponding to each of the moving objects. Then, it is desirable to execute the tracking process described in FIG. 11 for each cluster.

  Further, for example, each time the imaging of the partial areas dA_1 to dA_2m included in the observation area Ar illustrated in FIG. 1 is completed by the imaging apparatus EQ, the processor 21 is based on an image obtained by imaging of each partial area dA. A process for tracking the track of the moving object is executed.

  FIG. 11 shows tracking processing by the tracking device 10 shown in FIG. Each process of step S301 to step S306 and step S311 to step S314 illustrated in FIG. 11 is an example of a process included in the application program for the tracking process. In addition, each processing of step S301 to step S306 and step S311 to step S314 is executed by the processor 21 shown in FIG. Of the steps shown in FIG. 11, those equivalent to the steps shown in FIG. 4 are denoted by the same reference numerals.

  In step S301, the processor 21 receives information indicating observation points extracted from the partial area dA photographed by the photographing apparatus EQ via the general-purpose interface 24 shown in FIG.

  In step S302, the processor 21 is based on one of the wakes indicated by the hypothesis generated when the partial area dA photographed before the partial area dA photographed in the process of step S301. Find the gate area.

  In step S303, the processor 21 determines whether or not the observation point received in the process of step S301 is included in the gate region obtained in the process of step S302.

  When it is determined that the observation point received in step S301 is included in the gate region obtained in step S302 (Yes in step S303 (YES)), the processor 21 performs steps S304 and S311. Execute the process. On the other hand, when it is determined that the observation point received in the process of step S301 is not included in the gate region obtained in the process of step S302 (No determination in step S303 (NO)), the processor 21 performs the process in step S311. Execute the process.

  In step S304, the processor 21 generates a hypothesis that the moving body is observed at the extracted observation point, and uses the state vector of the first wake indicated by the generated hypothesis of the wake used for calculating the gate region. Estimation is performed using the state vector and observation point information.

  In step S311, the processor 21 generates a hypothesis that the moving object is not observed in the gate region, and uses the state vector of the second wake indicated by the generated hypothesis without using the information on the observation point. Estimated from the wake state vector used to calculate

  In step S305, the processor 21 determines the gate area for the entire track obtained in step S304 or step S311 for the partial area dA photographed before the partial area dA photographed in step S301. Determine whether or not.

  When there is a track for which the gate area has not yet been obtained (No at Step S305 (NO)), the processor 21 returns to Step S302 and repeats Steps S302 to S304 and Step S311. The processing of step S302 to step S304 and step S311 is repeatedly executed, and when the processing for all the wakes is completed (Yes determination in step S305), the processor 21 proceeds to the processing of step S312.

  In step S312, the processor 21 evaluates the likelihood of each hypothesis generated in the processes of steps S304 and S311 using the equations (1) to (3) described with reference to FIGS. The processor 21 stores the likelihood of each hypothesis obtained in the process of step S312 in a storage area provided in the memory 22 or the hard disk device 23, and the hypothesis corresponding to the child of each hypothesis subjected to the likelihood evaluation. Use when evaluating likelihood. Further, when evaluating the likelihood of each hypothesis, the processor 21 provides the obtained probability in the memory 22 or the hard disk device 23 when obtaining the probability that the moving object is not observed in the overlapping portion of the partial area dA and the gate area. It may be held in a separate storage area. Then, the processor 21 uses the “probability that no moving object is observed” held in the memory 22 or the hard disk device 23 as described in FIGS. 5 to 9 and equations (5) and (6). You may perform the process which correct | amends a degree.

  In step S313, the processor 21 detects a hypothesis having a likelihood greater than or equal to the second threshold based on the likelihood of each hypothesis obtained in the process of step S312, and the wake indicated by the detected hypothesis is displayed on the display device DSP. Process to be displayed. For example, the processor 21 reads the likelihood of each hypothesis newly obtained in the process of step S312 from the memory 22 or the like, and compares the read likelihood of each hypothesis with the second threshold value. Further, as a result of the comparison, the processor 21 passes the state vector of the wake indicated by the hypothesis having the likelihood equal to or greater than the second threshold to the display control unit 25 illustrated in FIG. The estimated position and the direction of motion are displayed on the display device DSP.

  In step S306, the processor 21 determines whether or not imaging of 2m partial areas dA included in the observation area Ar illustrated in FIG. 1 has been completed.

  If there is a partial area dA that has not yet been photographed (No at Step S306, NO), the processor 21 returns to Step S301 and repeats Steps S301 to S306 and Steps S311 to S313.

  When the processes for all the partial areas dA are completed by repeating the processes of steps S301 to S306 and steps S311 to S313 (Yes in step S306 (YES)), the processor 21 performs the process of step S314. Execute.

  In step S314, the processor 21 executes the pruning process described in FIG. 3 based on the likelihood of each hypothesis obtained by the process in step S312 in the process of capturing each partial area dA. Make a selection. Each of the wakes indicated by the hypothesis left by the pruning process is obtained by the processor 21 when the partial area dA of the observation area Ar is photographed next, and the past obtained by the tracking process up to the previous frame by the processor 21. Used as a wake.

  As described above, the tracking device 10 shown in FIG. 10 includes a wake indicated by a hypothesis that a moving body is observed in a plurality of partial areas dA, as described with reference to FIGS. Select a likely wake from a set of wakes that contain all possible wakes. As a result, the tracking device 10 shown in FIG. 10 can track a moving body with higher accuracy than in the past.

  Further, the tracking device 10 shown in FIG. 10 evaluates the likelihood of the hypothesis generated every time the partial area dA is photographed, and displays the wake indicated by the hypothesis having the likelihood equal to or higher than the second threshold value on the display device DSP. To display. Thereby, before the imaging of the partial area dA included in the observation area Ar is completed, information indicating the likely track of the moving body can be presented to the operator of the tracking device 10.

  From the above detailed description, features and advantages of the embodiment will become apparent. It is intended that the scope of the claims extend to the features and advantages of the embodiments as described above without departing from the spirit and scope of the right. Any person having ordinary knowledge in the technical field should be able to easily come up with any improvements and changes. Therefore, there is no intention to limit the scope of the inventive embodiments to those described above, and appropriate modifications and equivalents included in the scope disclosed in the embodiments can be used.

Regarding the above description, the following items are further disclosed.
(Appendix 1)
Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed A prediction unit for obtaining gate regions each including a predicted position of the moving body,
An extraction unit that extracts an observation point indicating a location having a feature amount equal to or higher than a first threshold included in the gate region obtained by the prediction unit, from each of the images of the plurality of partial regions obtained by photographing;
Each time each of the plurality of partial areas is photographed, a new wake is estimated using the observation point extracted by the extraction unit and the wake used to obtain the gate area including the observation point. An estimation unit;
A tracking unit that selects a wake having a higher likelihood than other wakes from a plurality of wakes of the moving object, including a new wake estimated by the estimating unit, as a likely wake of the moving object;
A tracking device characterized by comprising:
(Appendix 2) In the tracking device described in Appendix 1,
The estimation unit further estimates a new wake based on the wake used for obtaining the gate region without using the information of the observation point,
The tracking unit is
Each time each of the plurality of partial areas is photographed, the likelihood of the first wake, which is a wake estimated using the information on the observation point, by the estimation unit, is determined. The larger the area of the overlapping portion with the gate region obtained for the time when the image was taken, the larger the value, and the likelihood of the second wake which is the wake estimated by the estimator without using the observation point information. A tracking device, comprising: an evaluation unit configured to increase the degree as the area of a portion where the imaged partial region overlaps with the gate region obtained for the time when the partial region was imaged is smaller.
(Appendix 3) In the tracking device described in Appendix 2,
The evaluation unit is
In the evaluation of the likelihood of the first wake estimated by the estimation unit, the likelihood of the wake used for the estimation of the first wake and the moving body is observed in the captured partial region. The likelihood of the second wake estimated by the estimator, and the likelihood obtained for the wake used for the estimation of the second wake, and the captured partial area And a probability that the moving object is not observed.
(Appendix 4) In the tracking device described in Appendix 3,
The evaluation unit is
Probability that the moving object exists in a portion where the partial region and the gate region obtained for the time when the partial region overlaps are calculated in the calculation of the probability that the moving object is observed in the captured partial region. And a first value represented by a product of a constant indicating a probability that the moving object is observed in the region,
A tracking device, wherein a value obtained by subtracting the first value from 1 is used to calculate a probability that the moving object is not observed in the photographed partial area.
(Appendix 5) In the tracking device described in Appendix 3,
The evaluation unit is
As a probability that the moving object is not observed in the photographed partial area, a second value that is a value obtained by subtracting from 1 a constant indicating the probability that the moving object is observed in the area is used. Tracking device characterized by.
(Appendix 6) In the tracking device described in Appendix 3,
The evaluation unit is
When the likelihood of the first wake is evaluated, the mobile body used in the evaluation of the likelihood is observed with respect to the likelihood obtained for the wake used for the estimation of the first wake. A tracking device, wherein correction is performed to cancel out a change in the likelihood due to a probability of being not performed.
(Appendix 7)
In the tracking device according to attachment 2,
Furthermore,
An output unit for detecting a wake having a likelihood obtained by the evaluation unit equal to or greater than a second threshold in the process of photographing the plurality of partial areas, and outputting information representing the detected wake; Tracking device.
(Appendix 8)
Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed Each of the gate regions including the predicted position of the moving body,
From each of the images of the plurality of partial areas obtained by shooting, extract observation points indicating locations having a feature amount equal to or higher than a first threshold included in the gate area,
Every time each of the plurality of partial areas is photographed, a new wake is estimated using the observation points extracted from the gate area and the wake used to obtain the gate area,
Selecting a wake having a higher likelihood than the other wakes as a likely wake of the mobile body from a plurality of wakes of the mobile body including the estimated new wake;
The tracking method characterized by this.
(Appendix 9)
Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed Each of the gate regions including the predicted position of the moving body,
From each of the images of the plurality of partial areas obtained by shooting, extract observation points indicating locations having a feature amount equal to or higher than a first threshold included in the gate area,
Every time each of the plurality of partial areas is photographed, a new wake is estimated using the observation points extracted from the gate area and the wake used to obtain the gate area,
Selecting a wake having a higher likelihood than the other wakes as a likely wake of the mobile body from a plurality of wakes of the mobile body including the estimated new wake;
A tracking program that causes a computer to execute processing.

DESCRIPTION OF SYMBOLS 10 ... Tracking apparatus; 11 ... Prediction part; 12 ... Extraction part; 13 ... Estimation part; 14 ... Tracking part; 15 ... Output part; 141 ... Evaluation part; 142 ... Selection part; Hard disk device; 24 ... General-purpose interface; 25 ... Display control unit; 26 ... Input device; 27 Optical drive device; 28 ... Removable disk Ar ... Observation region; dA_1 to dA_2m ... Partial region; GP ... Image processing unit; DSP ... Display device

Claims (5)

Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed A prediction unit for obtaining gate regions each including a predicted position of the moving body,
An extraction unit that extracts an observation point indicating a location having a feature amount equal to or higher than a first threshold included in the gate region obtained by the prediction unit, from each of the images of the plurality of partial regions obtained by photographing;
Each time each of the plurality of partial areas is photographed, a new wake is estimated using the observation point extracted by the extraction unit and the wake used to obtain the gate area including the observation point. An estimation unit;
A tracking unit that selects a wake having a higher likelihood than other wakes from a plurality of wakes of the moving object, including a new wake estimated by the estimating unit, as a likely wake of the moving object;
A tracking device characterized by comprising: The tracking device according to claim 1,
The estimation unit further estimates a new wake based on the wake used for obtaining the gate region without using the information of the observation point,
The tracking unit is
Each time each of the plurality of partial areas is photographed, the likelihood of the first wake, which is a wake estimated using the information on the observation point, by the estimation unit, is determined. The larger the area of the overlapping portion with the gate region obtained for the time when the image was taken, the larger the value, and the likelihood of the second wake which is the wake estimated by the estimator without using the observation point information. A tracking device, comprising: an evaluation unit configured to increase the degree as the area of a portion where the imaged partial region overlaps with the gate region obtained for the time when the partial region was imaged is smaller. The tracking device according to claim 2,
Furthermore,
An output unit for detecting a wake having a likelihood obtained by the evaluation unit equal to or greater than a second threshold in the process of photographing the plurality of partial areas, and outputting information representing the detected wake; Tracking device. Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed Each of the gate regions including the predicted position of the moving body,
From each of the images of the plurality of partial areas obtained by shooting, extract observation points indicating locations having a feature amount equal to or higher than a first threshold included in the gate area,
Every time each of the plurality of partial areas is photographed, a new wake is estimated using the observation points extracted from the gate area and the wake used to obtain the gate area,
Selecting a wake having a higher likelihood than the other wakes as a likely wake of the mobile body from a plurality of wakes of the mobile body including the estimated new wake;
The tracking method characterized by this. Each time a plurality of partial areas obtained by dividing a predetermined area including a moving body are sequentially photographed in a time division manner, based on the track of the moving body estimated before the time when the partial area was photographed Each of the gate regions including the predicted position of the moving body,
From each of the images of the plurality of partial areas obtained by shooting, extract observation points indicating locations having a feature amount equal to or higher than a first threshold included in the gate area,
Every time each of the plurality of partial areas is photographed, a new wake is estimated using the observation points extracted from the gate area and the wake used to obtain the gate area,
Selecting a wake having a higher likelihood than the other wakes as a likely wake of the mobile body from a plurality of wakes of the mobile body including the estimated new wake;
A tracking program that causes a computer to execute processing.

Images