JP5063567B2 - Moving object tracking device - Google Patents

Description

  The present invention relates to a moving object tracking device that tracks a moving object that moves in a monitoring space.

  Conventionally, when a moving object moving in a monitoring space is tracked, a feature amount such as a texture or color of the object is stored as a template of the moving object, and has a feature amount similar to the template on an image obtained by capturing the monitoring space. The areas are sequentially associated as moving objects, and the associated areas are processed as moving objects.

  Furthermore, a process for preventing the association by mistake is also performed by performing the association only in the vicinity of the position of the current moving object predicted from the position of the past moving object.

JP 2008-250772 A

  By the way, in a monitoring space such as a store where an automatic cash transaction apparatus (ATM) is installed, a plurality of moving objects (people) may be arranged to form a queue. Since this matrix is formed by a large number of moving objects, a plurality of moving objects having similar characteristics may be included in the matrix. For this reason, there is a possibility that moving objects whose features are similar to each other are erroneously associated with each other, and the tracking accuracy of the moving object is lowered.

  In addition, when a matrix is formed, each moving object is almost stationary, so it is difficult to predict the current position from past movements, and it is difficult to narrow down moving objects to be associated from history information.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a moving object tracking device that improves the accuracy of associating moving objects.

  One aspect of the present invention is a moving object tracking device that tracks a moving object using a monitoring image obtained by sequentially imaging a monitoring space in which a matrix can be formed by the moving object, and the moving object is detected from the monitoring image. An object region extracting means for extracting an existing object region; and the object region and the moving object based on a similarity between an image in the object region and reference image information representing a feature on the image of each moving object. An association means for associating a plurality of moving objects with a matrix region in which a matrix is formed, a moving object associated with the matrix region as a matrix object, the matrix region, Matrix recognition means for storing, in a storage unit, matrix information including an entry order of the matrix object and the matrix object into the matrix, and detecting the object area leaving the matrix area as a departure area Comprising a disengagement region detection unit, wherein the correlating means, when said disengaging area detecting means detects the disengagement region, the entry order is moving object tracking apparatus for associating with priority the matrix object earliest.

  Here, the association unit determines the ambiguity of the association from the similarity distribution between the leaving area and the plurality of reference image information, and the entry order is the earliest in the leaving area determined to be ambiguous. It is preferable to associate the matrix object.

  Further, it is preferable that the leaving area detecting means detects the object area overlapping with the stored matrix area at a predetermined ratio as the leaving area.

  In addition, the association unit stores the position of the object region associated with the moving object in the storage unit as a movement locus of the moving object, and the matrix recognition unit refers to the movement locus and enters a stationary state. It is preferable that only the moving object associated with a certain object region is the matrix object.

  Further, the matrix recognition means stores the position of the matrix object with the earliest entry order included in the matrix information as the head position of the matrix, and the association means stores a predetermined distance range from the head position. It is preferable that the matrix object having the earliest entry order is preferentially associated with the departure area detected in the area.

  Further, the associating means enters the separation area detected between the matrix area and an equipment area on the monitoring image corresponding to an installation location of equipment provided for use of the moving object. The matrix object with the earliest order may be associated with priority.

  According to the present invention, it is possible to improve the specific accuracy of each moving object when tracking the moving object.

  As shown in FIG. 1, the moving object tracking device 1 according to the embodiment of the present invention includes an imaging unit 2, an operation unit 3, a storage unit 4, a control unit 5, and a display unit 6. The moving object tracking device 1 images a monitoring space, performs image processing on the monitoring image, and performs tracking processing of moving objects such as people and objects captured in the image. The imaging unit 2, the operation unit 3, the storage unit 4, the control unit 5, and the display unit 6 are connected so as to be able to transmit information to each other.

  In the present embodiment, an example of tracking a person's movement at an ATM corner that is a monitoring space will be described. However, the present invention is not limited to this, and the present invention can also be applied to cases such as detecting abnormal behaviors of articles such as commodities placed in distribution, passing cars, and the like.

  The imaging unit 2 is a so-called surveillance camera including an imaging element such as a CCD element or a C-MOS element, an optical system component, an analog / digital converter, and the like. The imaging unit 2 images the monitoring space at predetermined time intervals, and sequentially outputs the images (hereinafter referred to as monitoring images) to the control unit 5. Hereinafter, the predetermined time interval is referred to as time as a unit of time. In the present embodiment, the imaging unit 2 is arranged on the ceiling of the monitoring space with the optical axis directed vertically downward. Note that the imaging unit 2 may include a plurality of surveillance cameras.

  The operation unit 3 includes an input device such as a keyboard and a mouse for inputting information to the moving object tracking device 1. The operation unit 3 is used for inputting various settings by the administrator of the moving object tracking device 1.

  The storage unit 4 is configured by a memory device such as a ROM or a RAM. The storage unit 4 is connected so as to be accessible from the control unit 5. The storage unit 4 stores various programs and various data. These various programs and various data are read from the control unit 5 and used. The various data includes a monitoring image, object information 40, matrix information 41, and equipment area 42.

  The object information 40 is information indicating each feature of the moving object. The object information 40 is stored for each moving object together with an identifier for identifying each moving object, and is referred to in order to identify the moving object in the monitoring image. The object information 40 includes reference image information 400 that represents the appearance characteristics of the moving object, and a moving locus 401 that is a history (time-series data) of the position of the moving object and is generated as a result of tracking. The reference image information 400 can be image features such as a color histogram, shape, and size of a moving object. Note that a luminance histogram, texture, or the like may be used instead of or in addition to the color histogram.

  The matrix information 41 includes a matrix area that is an area in the monitoring image corresponding to an area where the matrix is formed in the monitoring space, an identifier of a moving object (hereinafter referred to as a matrix object) existing in the matrix area, and each matrix object Includes the entry order representing the order in which the entry into the matrix and the start position of the matrix. This matrix information 41 is stored for each matrix. 6 and 7 are examples showing the matrix information 41. FIG. 6 and FIG. 7 will be described later.

  The equipment area 42 is an area on the monitoring image corresponding to the installation location of equipment used for moving objects. The facility area 42 is, for example, a cash dispenser (CD), an automatic cash transaction apparatus (ATM), a bank window, or the like. The facility area 42 on the monitoring image can be set in advance by the administrator using the operation unit 3.

  The control unit 5 includes an arithmetic circuit such as a CPU, DSP, MCU, or IC. The control unit 5 is connected to the imaging unit 2, the operation unit 3, the storage unit 4, and the display unit 6 so as to be able to transmit information. The control unit 5 reads the program describing the processing in each unit such as the object region extraction unit 50, the association unit 51, the matrix recognition unit 52, the suspicious behavior detection unit 53 and the like from the storage unit 4 and executes it, thereby executing the computer. It functions as each means. The control unit 5 processes the monitoring image from the imaging unit 2 to track the moving object, and outputs an abnormal signal to the display unit 6 when detecting the suspicious behavior of the moving object by analyzing the movement trajectory obtained by the tracking process. To do.

  The display unit 6 includes sound output means such as a speaker and a buzzer that output a notification sound when an abnormal signal is input. Moreover, when an abnormal signal is input, a display unit that displays a movement locus or an image of a suspicious person related to the abnormality may be included.

  The moving object tracking device 1 may include an interface for connecting the computer to a network or a telephone line. In this case, the moving object tracking device 1 is connected to a center device (not shown) through an information transmission means such as a telephone line or the Internet so as to be able to transmit information. The moving object tracking device 1 and the center device transmit and receive information using a telephone number and a network address. The center device is a host computer installed in a monitoring center or the like where a monitor or the like stays. The center device includes a communication unit that receives report information from the moving object tracking device 1, a display unit that displays the report information, and the like.

  Hereinafter, the processing in the moving object tracking device 1 will be described with reference to the functional block diagram of FIG. 1 and the flowcharts of FIGS.

  The administrator of the moving object tracking device 1 turns on the moving object tracking device 1 after confirming that the monitoring space is unattended. As a result, processing as each unit and each means is started, and predetermined initialization is performed (S1). Further, the monitoring space is imaged by the imaging unit 2, and a monitoring image is output to the control unit 5. Thereafter, each time a new monitoring image is input from the imaging unit 2 to the control unit 5, the processes in steps S2 to S13 are repeated.

  When a new monitoring image is acquired (S1), the object region extraction means 50 of the control unit 5 extracts the object region by comparing the monitoring image with the background image (S2). Specifically, the monitoring image is compared with the background image, pixels that differ between the two images are detected, and the pixels detected adjacent to each other are collectively set as a change region, and a reference value having a predetermined size is set. The exceeding change area is extracted as an object area.

  The difference pixel is a pixel of a monitoring image in which a difference in luminance value or color component between the background image and the monitoring image is a predetermined value or more, or a luminance value or color component between the background image and the monitoring image. It is preferable that the pixel has a correlation value less than a predetermined value.

  However, since the background image is not acquired immediately after the power is turned on, the object region extracting unit 50 does not extract the object region, and stores the monitoring image acquired immediately after the power is turned on in the storage unit 4 as a background image. Further, at each subsequent time, the object region extraction means 50 averages the monitoring image with the background image for the region where the object region has not been extracted in order to cope with the change in the background due to the illumination variation or the like of the monitoring space. Update the image.

  The association means 51 of the control unit 5 calculates the predicted position of the current time from the movement trajectory 401 of each moving object (S3). The current predicted position is obtained by performing extrapolation processing on the movement trajectory 401 by applying a motion model such as a constant velocity motion model or a prediction filter such as a Kalman filter.

  Next, the associating means 51 performs matching processing using each reference image information 400 in each object region, and specifies a position in the object region where the similarity to the reference image information 400 is maximized (S4).

  The similarity is preferably calculated by combining the image similarity and the position similarity. For example, the similarity is obtained using Equation (1).

[Equation 1]
Similarity = α × image similarity + (1−α) × position similarity (1)
However, 0 <α <1.

  For the image similarity, a candidate position is set in the object area, an area having the shape and size indicated by the reference image information 400 is set based on the candidate position, and the color histogram extracted from the monitoring image in the area and the reference image The correlation function is obtained by applying a correlation function having a larger value as the difference from the color histogram of the information 400 is smaller. The position similarity is obtained by applying a correlation function that becomes larger as the distance between the predicted position obtained from the movement locus 401 stored as the object information 40 and the candidate position is smaller. The associating means 51 repeats the calculation of the similarity while shifting the candidate position until the maximum similarity is obtained, and specifies the candidate position where the maximum similarity is calculated. Thus, the similarity and position are calculated for all combinations of the object area and the moving object.

  Then, the associating unit 51 temporarily associates the moving object with the object region having the maximum similarity for each moving object and the similarity is equal to or greater than the predetermined identification value L1 (S6). Further, the current position of each moving object is determined as the candidate position for which the maximum similarity is calculated. As a result, in the matrix area, a plurality of matrix objects are associated with the same matrix area, and the positions of the plurality of matrix objects are determined in the same matrix area.

  Here, the object region may include an object region that has left the matrix region (hereinafter referred to as a leaving region).

  In many cases, a matrix formed from a large number of matrix objects includes a plurality of matrix objects having similar appearances (such as human clothes). Indicates. Further, since the matrix objects are close to each other in the matrix and their movements are stationary or slow, the position similarity of the matrix objects with respect to the leaving region is also side by side. For this reason, it is often the case that correct association cannot be made with respect to the separation region only by the similarity of images and positions.

  Therefore, the association unit 51 includes a leaving area detection unit 510 that detects an object area that leaves the matrix area, and performs processing for appropriately correcting the association with the detected leaving area in the association result in step S6. Perform (S7).

  The process of step S7 is demonstrated with reference to the flowchart of FIG. The leaving area detecting means 510 of the associating means 51 detects an object area whose degree of overlap with the matrix area at the previous time is within the leaving determination range as a leaving area (S100).

  Due to the nature of leaving from the matrix, the leaving area is detected as an area that partially overlaps the matrix area at the previous time and the remaining part protrudes from the matrix area at the previous time. Accordingly, the leaving area detection unit 510 calculates the degree of overlap between the matrix area of the previous time and each object area using Equation (2), and sets the object area whose degree of overlap is within a predetermined leaving determination range as the leaving area. To detect.

[Equation 2]
Degree of overlap = area overlapping with matrix area / area of object area (2)

  For example, the leaving determination range is set to (0, 0.5). That is, when the upper limit of 0.5 is set as the leaving determination threshold TH, the leaving area detection unit 510 matrixes at a ratio equal to or less than the leaving determination threshold TH. An object region that overlaps with the region (less than half of the object region overlaps with the matrix region) is detected as a departure region, and overlaps with the matrix region at a rate exceeding the separation determination threshold TH (more than half of the object region) It is determined that the object region (the region overlaps with the matrix region) is included in the matrix region, and the object region with the overlap degree 0 (not overlapping with the matrix region) is determined to be irrelevant to the matrix. A lower limit value (for example, 0.1) greater than 0 may be set so that the object area of the matrix is not erroneously detected as a leaving area due to movement.

  In another example, an object area associated with a matrix object alone may be detected as a leaving area. However, in the method of detecting the separation area based on the degree of overlap, even if the gap between the matrix objects is imaged, a part of the matrix is not erroneously detected as the separation area due to the influence of the gap. Excellent in terms. In particular, in a matrix having a large angle between the matrix direction and the imaging direction, gaps are easily imaged, and the effect is great.

  Next, the associating unit 51 executes a loop process of steps S101 to S107 for the leaving area.

  In the separation area, there are a moving object that has slipped out from the head in a line and a moving object area that has slipped out without waiting. In order to distinguish these, the associating means 51 compares the position of the head position of the matrix where the leaving area is detected with the position of the leaving area to determine the positional relationship (S102), and the leaving area is a distance range from the head position. If it is in the vicinity of the threshold value D, the process proceeds to step S104, assuming that it is an escape from the head, and otherwise, the process proceeds to step S107, assuming that it is a void (S103). In another example, the associating means 51 makes a break-out from the head if the leaving area is between the matrix area and the equipment area 42, and otherwise makes a hollow.

  FIG. 8 is an example schematically showing the processing of S100 to S103. An area 80 captured as a monitoring image shows an equipment area 42, an object area 81 extracted at the current time, and a matrix area 82 recognized at the previous time. The x (cross) mark represents the head position of the matrix, and the circle 83 represents the distance range threshold D.

  In FIG. 8, the overlap between the object region 81-2 and the matrix region 82 is 0.2, which is within the departure determination range, and the object region 81-2 is detected within the distance range threshold D. -2 is a departure area that has escaped from the top of the matrix. On the other hand, since the overlapping degree of the object area 81-1 is 0 and the overlapping degree of the object area 81-3 is 1, it is not detected as a separation area.

  As the distance range threshold D, it is preferable to set a value that the moving object can take as the initial speed. The initial speed is the speed at which a moving object that has been stationary starts to move, and specifically corresponds to the moving distance per time when the moving objects that are arranged in a matrix exit the matrix.

  The association unit 51 determines the ambiguity of the association with the departure area from the similarity distribution calculated between the departure area and the plurality of reference image information. Here, the ambiguity refers to a property indicating the certainty degree of the result of the association between the leaving area and the reference image information. The association determined to be ambiguous has a low accuracy of association based on the similarity, and may be erroneously associated.

  More specifically, the associating means 51 is ambiguous when the maximum similarity between the leaving area and the plurality of reference image information is less than the identified value L2, in other words, when all the similarities are less than the identified value L2. It is determined to have The identification value L2 is set to a value larger than the identification value L1 used for provisional association in step S6.

  FIG. 9 is a diagram for explaining the ambiguity determination. Circles indicate the similarity between the separation region and the reference image information for each matrix object. In FIG. 9A, the ambiguity is not determined because the maximum similarity is equal to or greater than the identification value L2, and in FIG. 9B, the ambiguity is determined because the maximum similarity is less than the identification value L2.

  Further, the association unit 51 may determine that there is ambiguity when there are a plurality of similarities equal to or greater than the identification value L1 for the leaving region. In FIG. 9 (c), the degree of ambiguity is not determined because the degree of similarity is equal to or higher than the identification value L1, and in FIG. 9 (d), the degree of ambiguity is not higher than two because the degree of similarity is equal to or higher than the identification value L1. Determined.

  Further, the association unit 51 may be ambiguous when the difference d between the top two similarities among the similarities calculated for the leaving area is less than a predetermined value R. In FIG. 9E, the ambiguity is not determined because the difference d is greater than or equal to the value R. In FIG. 9F, the ambiguity is determined because the difference d is less than the value R.

  If the ambiguity is not determined, the result of the temporary association between the object region and the moving object in step S6 is used as the final association as it is (YES in S105). On the other hand, when the ambiguity is determined (NO in S105), the associating unit 51 re-associates the moving object associated with the leaving area with the matrix object with the earliest entry order in the matrix in which the leaving area is detected ( S106). In this case, the matrix object associated with the leaving area in step S6 is associated with the matrix area again. The position of the matrix object that has been re-associated is corrected with reference to the position specified in step S4.

  Since the matrix has the property of leaving the matrix in order starting with the objects that are arranged first, multiple similarities can be obtained by correcting the matrix object with the highest entry order in the leaving area as described above. Even when the matrix object is included in the matrix, correct association is possible, and the accuracy of specifying each moving object is increased.

  At this time, by performing correction on the leaving area for which the ambiguity of the association is determined and not performing correction on the leaving area that is not so, the association based on the similarity can be taken into account, and each moving object is identified. The accuracy increases further.

  In addition, correction is performed for a separation area detected within a predetermined range from the head position, or a separation area detected between the matrix area and the facility area 42, and correction is not performed for other separation areas. In this way, it is possible to prevent an error that associates the earliest matrix object with a hollow leaving area.

  After the association, the association unit 51 additionally stores the position of the detected moving object in the movement locus 401 of the object information 40, and replaces the reference image information 400 with the image feature of the associated object candidate area (S8). . Note that the reference image information 400 is not updated for matrix objects.

  Note that a moving object that has not been associated with any object region for a predetermined period has disappeared from the monitoring space, and the object information 40 is deleted and excluded from the tracking target. Also, assuming that the object area that is not associated with any moving object is that of a newly appearing moving object, the image feature and position of the object area are newly stored in the storage unit 4 as object information 40, and the tracking target Add to.

  Next, the matrix information 41 is generated in the matrix recognition means 52 of the control unit 5 (S9). With reference to FIG.4 and FIG.5, the process of step S9 is demonstrated.

  First, the matrix recognition means 52 performs a loop process of steps S200 to S203 on the stored matrix information 41, and updates the matrix information 41 associated with the departure. That is, the matrix recognition unit 52 refers to the association result, and if there is a matrix object associated with the leaving area (YES in S201), deletes information related to the matrix object from the matrix information 41 (S202). . Further, the number of matrix objects included in the matrix information 41 is counted, and the matrix information 41 in which the number of matrix objects is less than 2 is deleted (YES in S203, S204). Note that there are a case where a matrix object missing from the head is associated with the leaving area and a case where a missing matrix object is associated with.

  Even if the matrix object associated with the departure area is deleted, the matrix recognition means 52 determines that the matrix object having the number of matrix objects of 2 or more matches the matrix information 41 with the matrix at the current time. The matrix area is replaced with a circumscribed rectangle of the object area extracted inside (S205). Further, the head position is replaced with the position associated with the matrix object with the earliest entry order (S206).

  In this embodiment, the matrix area is processed as a circumscribed rectangular area for simplification of processing, but the outer shape of a set of object areas of matrix objects constituting the matrix may be processed as a matrix area.

  If there is no matrix object associated with the leaving area (YES in S201), the processes in steps S202 to S206 are skipped.

  It is determined whether or not the processing has been completed for all the matrices. If the processing in steps S200 to S206 has been performed for all the matrices, the processing proceeds to step S208. If not, the processing proceeds to step S208. The process is returned to S200 and the process for the next matrix is performed.

  Next, the matrix recognizing means 52 refers to the association result, selects an object area (multiplexed area) associated with a plurality of moving objects (S208), and performs steps S209 to S209 for the selected multiplexed area. The loop processing of S226 is performed, and new matrix information 41 is generated and matrix information 41 is updated.

  The matrix recognition means 52 determines whether or not the multiplexing area is the same as the existing matrix from the degree of overlap between the multiplexing area and the matrix information recognized one time ago (S210). If it is determined that they are not identical, there is a possibility of a new matrix. In this case (NO in S211), the matrix recognition means 52 refers to the movement trajectory 401 and determines whether or not each moving object associated with the multiplexed region is in a stationary state (S212).

  In the present embodiment, when the position of the object area of each moving object is in a state in which it does not substantially change without exceeding a predetermined moving range for a predetermined stationary time Ts or longer, the moving object is in a stationary state. It is determined that

  If it is determined that a plurality of moving objects are in a stationary state (YES in S213), matrix recognition means 52 sets the multiplexed area as a new matrix area, and sets each moving object associated with the multiplexed area as a matrix object. Recognize. The matrix recognition means 52 assigns the entry order to the recognized matrix objects in the order of the longer stationary state (S214), and calculates the circumscribed rectangle of the multiplexed area as the matrix area (S215). Further, the position where the matrix object with the earliest entry order is associated is obtained as the head position (S216), the recognized matrix objects, the entry order assigned to these matrix objects, the calculated matrix area and the obtained matrix area are obtained. The head position is stored in the storage unit 4 as new matrix information 41 (S217).

  On the other hand, when it is determined that the multiplexed region is the same as the existing matrix (YES in S211), the matrix recognition unit 52 refers to the matrix information 41 of the existing matrix and moves the moving object associated with the multiplexed region. A moving object (new multiplexed object) that is not stored as a matrix object is detected from among the objects (S218), and it is determined by referring to the movement trajectory 401 whether or not the new multiplexed object is in a stationary state (S219). YES, S220).

  If it is determined that it is in a stationary state (YES in S221), the matrix recognition means 52 recognizes the new multiplexed object as a matrix object, adds it to the matrix information 41 (S222), and enters the existing matrix object as the entry order. A value obtained by adding 1 to the maximum value of the entry order is assigned (S223). Further, the matrix recognition means 52 replaces the matrix area with a circumscribed rectangle for all object areas associated with the matrix objects constituting the same matrix in order to add the newly entered matrix object area to the matrix area (S224), The head position is replaced with the coordinates associated with the matrix object with the earliest entry order (S225).

  Note that the multiplexed region occurs not only when a matrix is generated, but also when an image of a moving object that is continuously moving intersects on the monitoring image. Accordingly, if the new multiplexed object is not in a stationary state (NO in S221), there is a possibility that the moving object intersects while moving with each other, so that the matrix information 41 is updated (S222 to S225). Skip. Also, when a new multiplexed object is not detected (NO in S219), updating of matrix information 41 (S219 to S225) is skipped.

  FIG. 6 and FIG. 7 illustrate the state of matrix recognition. 6 and 7 show an area 70, an equipment area 42, an object area 72, a matrix area 73, and matrix information 41 that are captured as monitoring images at times t1 to t7. In the vicinity of each object region 72, identifiers #A to #D indicating corresponding moving objects are attached. In addition, the x mark represents the head position of the matrix, Q represents the coordinates of the head position, P1 to P4 represent the coordinates of the vertices of the matrix area 73, and the arrows represent a part of the movement locus 401.

  At time t1, the moving object #A starts to stand still, and at time t2, the moving object #B starts to stop near the moving object #A, and these moving objects are associated with the object region 72-2. Then, at time t3 when the stationary time Ts has elapsed from time t2, the moving objects #A and #B are both associated with the object region 72-3, and the respective moving objects #A and #B are in a stationary state. Therefore, the matrix recognition means 52 determines that the object region 72-3 is a matrix region and the moving objects #A and #B are matrix objects.

  The matrix recognition means 52 obtains the circumscribed rectangle Rect (P1, P2) of the object area 72-3 as the matrix area 73-3, obtains the position of the moving object #A as the head position Q, and moves the moving object # having a longer stationary state # The entry order No. 1 is assigned to A, the entry order No. 2 is assigned to the moving object #B, and the matrix information 41-3 including these is stored in the storage unit 4.

  At time t4, the moving object #C overlaps the matrix area 73-4, and the moving area #C is associated with the object area 72-4 in addition to the moving objects #A and #B. However, at this time point, the moving object #C is not determined to be stationary, and the moving object #C is not determined as a matrix object. The contents of the matrix information 41-4 are not updated from the previous time.

  The moving object #C is associated with an object region 72-5b different from the object region 72-5a at time t5 before the stationary time Ts elapses from time t4. That is, it is determined that the moving object #C has not entered the matrix but only crossed the matrix area at time t4, and the contents of the matrix information 41-5 are not updated from the previous time.

  At time t6, the object region #D starts to stand still in the vicinity of the moving objects #A and #B. At time t7 when the stationary time Ts has elapsed from time t6, the moving object #A, #B, and #D are associated with the object region 72-7, and it is determined that the moving object #D is in a stationary state. The matrix recognition means 52 obtains the circumscribed rectangle Rect (P3, P4) using the object region 72-7 as the matrix region, updates the head position Q, attaches the entry order number 3 to the moving object #D, and sets the matrix information 41. Update -7.

  Thus, the matrix recognition process is performed by the matrix recognition means 52, and each matrix information 41 is updated.

  Next, the suspicious behavior detection means 53 of the control unit 5 analyzes the movement trajectory 401 of each moving object to determine whether or not it corresponds to the suspicious behavior (S10), and detects the suspicious behavior (YES in S11). ), An abnormal signal including suspicious image information is generated and output to the display unit 6 (S12).

  Note that suspicious behavior can be detected by applying an existing method. For example, a normal action pattern that collects normal action patterns of moving objects is compared with the movement trajectory 401 of each moving object, and any moving object that deviates from the normal action pattern is detected as a suspicious action. do it. Further, if there is a moving object staying in the monitoring space for a predetermined time or more, it may be detected as suspicious behavior.

  The display unit 6 to which the abnormal signal is inputted outputs a warning by the sound output means, and displays the suspicious image information on the monitor device. The suspicious image information is, for example, a still image or a moving image obtained by cutting out the vicinity of the position of the moving object indicated by the movement trajectory 401 in which the suspicious behavior is detected, or the movement trajectory 401 in which the suspicious behavior is detected in the monitoring image. It may be a still image or a moving image in which figures and characters that emphasize the position of the moving object are combined. As a result, the monitoring person grasps the necessity of countermeasures such as guidance or exclusion based on the suspicious image information, and if necessary, promptly approaches the monitoring space to perform the countermeasures. When the above process is completed, the process returns to step S1 again.

  Further, in the above embodiment, the association unit 51 preferentially associates the moving object with the earliest entry order with the leaving area when the ambiguity is determined. In another embodiment, preferential association is performed by assigning a higher weight to the degree of similarity in the order of entry into the matrix. In this case, the similarity calculation formula may be, for example, the following formula.

[Equation 3]
Similarity = α × image similarity + β × position similarity + (1−α−β) · w
w = (number of matrix objects−entry order) / (number of matrix objects)
However, 0 <α <1, 0 <β <1, α + β = 1.

  As described above, according to the present embodiment, it is possible to improve the accuracy of specifying each moving object constituting a matrix and identifying each moving object that leaves the matrix when tracking the moving object.

  Further, in the present embodiment, the mode in which the function of each unit of the moving object tracking device 1 is realized by one computer has been described, but the present invention is not limited to this. The functions of each part of the moving object tracking device 1 can be realized by controlling a general computer by a program, and the functions of these devices may be appropriately combined and processed by one computer. May be distributed by a plurality of computers connected by a network or the like.

It is a functional block diagram which shows the structure of the moving object tracking apparatus in embodiment of this invention. It is a flowchart which shows the process in the moving object tracking apparatus in embodiment of this invention. It is a flowchart which shows the matching correction | amendment process in the moving object tracking device in embodiment of this invention. It is a flowchart which shows the matrix recognition process in the moving object tracking apparatus in embodiment of this invention. It is a flowchart which shows the matrix recognition process in the moving object tracking apparatus in embodiment of this invention. It is a figure explaining the matrix recognition process in embodiment of this invention. It is a figure explaining the matrix recognition process in embodiment of this invention. It is a figure explaining the correlation correction | amendment process in embodiment of this invention. It is a figure explaining the determination process of the ambiguity in embodiment of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Moving object tracking apparatus, 2 Imaging part, 3 Operation part, 4 Storage part, 5 Control part, 6 Display part, 40 Object information, 41 Matrix information, 42 Equipment area | region, 50 Object area extraction means, 51 Correlation means, 52 Matrix recognition means, 53 Suspicious behavior detection means, 70 area, 73 matrix area, 80 area, 81 object area, 82 matrix area, 83 distance range threshold D, 400 reference image information, 401 movement locus, 510 departure area detection means.

Claims (6)

A moving object tracking device that tracks a moving object using monitoring images obtained by sequentially imaging a monitoring space in which a matrix can be formed by the moving object,
Object region extraction means for extracting an object region where the moving object is present from the monitoring image;
Association means for associating the object region with the moving object based on the similarity between the image in the object region and reference image information representing a feature on the image of each moving object;
The object region associated with the plurality of moving objects is a matrix region in which a matrix is formed, the moving object associated with the matrix region is a matrix object, and the matrix region, the matrix object, and the matrix object Matrix recognition means for storing matrix information including the order of entry into the matrix in the storage unit;
A leaving area detecting means for detecting the object area leaving the matrix area as a leaving area, and
The moving unit tracking apparatus according to claim 1, wherein when the leaving area detecting unit detects a leaving area, the associating unit preferentially associates the matrix object with the first entry order.   The association means determines the ambiguity of association from the similarity distribution between the leaving area and the plurality of reference image information, and the matrix object having the first entry order in the leaving area determined to be ambiguous The moving object tracking device according to claim 1, wherein:   3. The moving object tracking device according to claim 1, wherein the leaving area detecting unit detects the object area that overlaps the stored matrix area at a predetermined ratio as the leaving area. 4. . The association unit causes the storage unit to store a position in the object region associated with the moving object as a movement locus of the moving object,
4. The moving object tracking apparatus according to claim 1, wherein the matrix recognition unit refers to only the moving object that is in a stationary state with reference to the moving locus as the matrix object. 5. The matrix recognition means includes the position of the matrix object with the earliest entry order included in the matrix information as the head position of the matrix, and stores it.
5. The association unit according to claim 1, wherein the association unit preferentially associates the matrix object having the first entry order with the departure area detected within a predetermined distance range from the head position. The moving object tracking apparatus as described in any one.   The associating means has the entry order in the separation area detected between the matrix area and the equipment area on the monitoring image corresponding to the installation location of equipment to be used for the moving object. 5. The moving object tracking apparatus according to claim 1, wherein the first matrix object is associated with priority.

Images