Traffic volume counting method and traffic volume counting system

Abstract

The present invention enables highly accurate counting the number of vehicles traveling on a road regardless of the occurrence of vehicle tracking errors.　A vehicle tracking unit 51 carries out a process for acquiring the trajectory of a vehicle by analyzing a video from a camera 10. A vector acquisition unit 61 carries out a process for acquiring a movement direction vector indicating the direction of an end point position with respect to a start point position related to the trajectory of the vehicle. A direction similarity degree calculation unit 64 carries out a process for calculating the degree of similarity, with respect to a reference angle of view of the camera 10, between the movement direction vector and a preset road direction vector. When the absolute value of the degree of similarity is greater than a prescribed threshold value, a direction determination unit 65 carries out a process for determining that the vehicle is traveling in the direction along the road direction vector, and counting the number of vehicles.

Description

Traffic volume counting method and traffic volume counting system

[0001] The present invention relates to a traffic counting method and a traffic counting system for counting the number of vehicles traveling on a road.

[0002] Traditionally, traffic counting has been done to understand road usage and congestion by analyzing camera footage of vehicles traveling on highways, public roads, etc. Traffic counting not only requires detecting vehicles present on the screen at any given time, but also analyzing whether they are moving in an uphill or downhill direction on the road.

[0003] Fig. 1 shows an example of the configuration of a conventional traffic volume counting system. The traffic volume counting system in the figure includes a camera 10, a transmission unit 20, a count recording unit 30, and a traffic volume count processing unit 40. The camera 10 is installed to capture images of a road on which traffic volume counting is to be performed. The transmission unit 20 transmits video data output from the camera 10 to the traffic volume count processing unit 40. The transmission unit 20 can be implemented using any communication method, such as wired communication using a LAN cable or coaxial cable, or wireless communication using Wi-Fi or a mobile phone network.

[0004] The traffic volume count processing unit 40 includes a vehicle tracking unit 41, a trajectory holding unit 42, a virtual line holding unit 43, and an intersection determination unit 44. The traffic volume count processing unit 40 may be mounted on an edge device installed near the site, or may be mounted on a server such as a cloud server.

[0005] The vehicle tracking unit 41 analyzes successive images (images of each frame) obtained from the video data of the camera 10, detects the position and size (Bbox) of the vehicle in the image, and acquires the trajectory of each vehicle within the field of view by comparing these between frames and linking the same vehicle. The vehicle tracking unit 41 can be realized, for example, by combining the publicly known object detection AI model described in Non-Patent Document 1 with the publicly known tracking techniques described in Non-Patent Document 2 and Non-Patent Document 3.

[0006] The trajectory storage unit 42 stores data on the trajectory of each vehicle acquired by the vehicle tracking unit 41 (e.g., data representing the vehicle position on a frame-by-frame basis). The virtual line storage unit 43 stores data on virtual lines, which are line segments virtually set on roads within the field of view of the camera 10 before operation begins. Generally, virtual lines are set in a direction perpendicular to the extension direction of the road (i.e., the width direction of the road). The intersection determination unit 44 determines in which direction (uphill or downhill) the trajectory of each vehicle in the trajectory storage unit 42 crosses the virtual line in the virtual line storage unit 43, and counts the number of vehicles by direction. The number of vehicles by direction counted by the intersection determination unit 44 is stored in the count recording unit 30.

[0007] Redmon, Joseph, and Ali Farhadi., "YOLOv3: An Incremental Improvement.," arXiv:1804.02767Wang, Zhongdao, et al., "Towards Real-Time Multi-Object Tracking.," ECCV2020Zhang, Yifu, et al., "ByteTrack: Multi-Object Tracking by Associating Every Detection Box.," ECCV2022

[0008] The accuracy of counting in the above method depends on the setting of the virtual line, as shown in Figures 2A and 2B. Figure 2A shows an example where the virtual line is set inappropriately, while Figure 2B shows an example where the virtual line is set appropriately. In Figures 2A and 2B, the solid arrows indicate the actual vehicle trajectories, and the dashed arrows indicate the vehicle trajectories acquired by the vehicle tracking unit 41. In this example, when vehicle 1 traveling in the uphill direction and vehicle 2 traveling in the downhill direction approach each other, the vehicle tracking unit 41 makes a tracking error due to vehicle confusion or misrecognition. Specifically, referring to Figure 2A, it is determined that vehicle 1 makes a sudden turn below the virtual line. At this time, the disappearance of vehicle 2 and the new appearance of vehicle 1 are also detected above the virtual line.

[0009] In a situation where such a tracking error occurs, in the case of Figure 2A where the virtual line is set improperly, none of the vehicle's trajectories cross the virtual line, so no vehicle is counted in either the up or down direction. On the other hand, in the case of Figure 2B where the virtual line is set appropriately, the trajectory of vehicle 1 crosses the virtual line in both the up and down directions, so one vehicle is counted in each direction. Therefore, in order to properly count traffic volume in a conventional system, it can be understood that it is important to set the virtual line in a location where vehicle tracking errors (errors in vehicle trajectories) are unlikely to occur (or where the impact of tracking errors is small).

[0010] However, in the past, workers manually set virtual lines according to the camera's installation position, and the accuracy of traffic counts depended on the worker's level of skill. Another problem was that the virtual lines had to be re-set whenever the camera's angle of view changed. This was particularly problematic for cameras with pan-tilt-zoom (PTZ) functionality, where it was difficult to re-set the appropriate virtual lines every time the angle of view changed due to changes in the camera's orientation or zoom.

[0011] The present invention has been made in consideration of the above-described conventional circumstances, and aims to enable the number of vehicles traveling on a road to be counted with high accuracy without being affected by vehicle tracking errors, etc.

[0012] In order to achieve the above object, a traffic volume counting method according to one aspect of the present invention is configured as follows: That is, a traffic volume counting method for counting the number of vehicles based on video captured by a camera capturing a road includes the steps of: analyzing the video captured by the camera to acquire vehicle trajectories; acquiring a movement direction vector indicating the direction of an end point relative to a start point of the vehicle trajectory; calculating a similarity between a road direction vector preset for a reference angle of view of the camera and the movement direction vector; and, if the absolute value of the similarity is greater than a predetermined threshold, determining that the vehicle is traveling in a direction along the road direction vector and counting the number of vehicles.

[0013] Here, in the above traffic volume counting method, when only one road direction vector is set for one road, the process of counting the number of vehicles may determine whether the vehicle is traveling in the forward or reverse direction of the road direction vector depending on whether the similarity is positive or negative, and count the number of vehicles by direction.

[0014] Furthermore, in the above traffic volume counting method, an inter-frame movement vector representing the direction of movement of the vehicle between frames may be obtained based on the vehicle's trajectory, and if the absolute value of the similarity between the road direction vector and the inter-frame movement vector is smaller than the threshold value, or if the sign of the inter-frame movement vector is reversed compared to the immediately preceding frame, the vehicle's trajectory related to the inter-frame movement vector may be excluded from the counting target.

[0015] In the above traffic volume counting method, the similarity may be calculated after a vector conversion process corresponding to a change in the angle of view of the camera is performed on the road direction vector or the movement direction vector.

[0016] A traffic volume counting system according to another aspect of the present invention is configured as follows: That is, a traffic volume counting system that counts the number of vehicles based on video captured by a camera of a road includes a vehicle tracking unit that analyzes the video captured by the camera to acquire vehicle trajectories, a vector acquisition unit that acquires a movement direction vector indicating the direction of an end point relative to a start point of the vehicle trajectory, a direction similarity calculation unit that calculates a similarity between a road direction vector preset with respect to a reference angle of view of the camera and the movement direction vector, and a direction determination unit that determines that the vehicle is traveling in a direction along the road direction vector when the absolute value of the similarity is greater than a predetermined threshold, and counts the number of vehicles.

[0017] According to the present invention, it is possible to count the number of vehicles traveling on a road with high accuracy, regardless of the occurrence of vehicle tracking errors.

[0018] FIG. 1 is a diagram showing an example of the configuration of a conventional traffic volume counting system. FIG. 2 is a diagram showing an example of a case where virtual lines are set inappropriately. FIG. 3 is a diagram showing an example of a case where virtual lines are set appropriately. FIG. 4 is a diagram showing an example of the configuration of a traffic volume counting system according to an embodiment of the present invention. FIG. 5 is a diagram showing an example of the configuration of a movement direction determination unit in the traffic volume counting system of FIG. 3. FIG. 6 is a diagram showing an example of a road direction vector. FIG. 7 is a diagram showing examples of inter-frame movement vectors and movement direction vectors. FIG. 8 is a diagram showing an example of a flowchart related to compensation of IDSW by a tracking result correction unit. FIG. 9 is a diagram showing an example of a vehicle ID before tracking result correction is performed. FIG. 10 is a diagram showing an example of a vehicle ID after tracking result correction is performed.

[0019] An embodiment of the present invention will be described with reference to the drawings. FIG. 3 shows an example of the configuration of a traffic volume counting system according to an embodiment of the present invention. The traffic volume counting system shown in FIG. 3 is compatible with conventional traffic volume counting systems, and can be operated by replacing the traffic volume counting processing unit 40 shown in FIG. 1 with a traffic volume counting processing unit 50 without changing the other components. However, for the purposes of processing described below, the traffic volume counting processing unit 50 receives not only the video data output from the camera 10 but also information related to changes in the angle of view of the camera 10 (internal parameters related to the orientation and focal length of the camera 10) via the transmission unit 20.

[0020] The traffic volume count processing unit 50 has a vehicle tracking unit 51, a movement direction determination unit 52, and a tracking result correction unit 53. The traffic volume count processing unit 50 is realized by, for example, a computer equipped with hardware resources such as a processor and memory, and is configured to implement functions and processing related to the present invention by reading a predetermined program from the memory and executing it with the processor. The traffic volume count processing unit 50 may be installed in an edge device installed near the site, or may be installed on a server such as a cloud server.

[0021] The vehicle tracking unit 51 performs object tracking processing by analyzing successive images (images of each frame) obtained from the video data of the camera 10 in a manner similar to that of the vehicle tracking unit 41 in FIG. 1 to detect vehicles and link each vehicle between frames. During the object tracking processing, a unique number (vehicle ID) is assigned to each vehicle. The vehicle tracking unit 51 can be realized using any tracking technology. For example, a tracking technology that takes into account the prediction of position information and the appearance of the vehicle, as in Non-Patent Document 2, or a tracking technology that uses a simple algorithm but has high accuracy, as in Non-Patent Document 3, may be used.

[0022] The vehicle tracking unit 51 outputs vehicle trajectory data (e.g., data representing the vehicle position in frame units) for each vehicle ID as a detection result of the object tracking process. At this time, the vehicle tracking unit 51 may output vehicle trajectory data obtained using all detection results, or may output vehicle trajectory data obtained using detection results excluding those outside the upper and lower limits of a preset detection size.

[0023] The movement direction determination unit 52 directly counts the number of vehicles traveling on the road using vectors obtained from the trajectory of each vehicle, instead of the conventional counting process that uses the intersection of a virtual line and the trajectory of each vehicle. The movement direction determination unit 52 also has a function of performing inverse conversion of changes in the angle of view of the camera 10 due to PTZ, so that the counting process can be continued without resetting even if the angle of view of the camera 10 changes.

[0024] Here, an overview of the vectors handled by this system will be described with reference to FIGS. 5A and 5B. FIG. 5A shows an example of a road direction vector. The road direction vector represents the extension direction of the road and connects any two points on the same lane in a specific direction (uphill or downhill). For example, as shown in FIG. 5A, the road direction vector can be calculated from two points, the start position P1 and the end position P2 of the road on the screen. The road direction vector can be obtained simply by the operator specifying the start and end points of the road on the screen displaying an image captured at the reference angle of view (initial setting angle of view) of the camera 10 during initial setup, and the data is stored (memorized) in the road direction storage unit 63. Note that if the road is bidirectional, two road direction vectors may be set separately for the uphill and downhill directions, or a single road direction vector may be set in one direction passing through any position in the width direction of the road (e.g., the center of the road).

[0025] 5B shows examples of inter-frame motion vectors and motion direction vectors. The inter-frame motion vector is a vector that indicates the direction of motion of the vehicle in each frame of the camera image. The motion direction vector is a vector that indicates the direction of the end position relative to the start position of the vehicle trajectory obtained by the vehicle tracking unit 51. The motion direction vector is equal to the sum of a series of inter-frame motion vectors obtained for the vehicle.

[0026] The movement direction determination unit 52 performs processing to determine the movement direction of the vehicle based on the trajectory of the vehicle obtained by the vehicle tracking unit 51. Fig. 4 shows an example of the configuration of the movement direction determination unit 52. The movement direction determination unit 52 has a vector acquisition unit 61, a vector conversion unit 62, a road direction storage unit 63, a directional similarity calculation unit 64, and a direction determination unit 65.

[0027] The vector acquisition unit 61 acquires inter-frame movement vectors and movement direction vectors of each vehicle based on the vehicle trajectory acquired for each vehicle ID by the vehicle tracking unit 51. The inter-frame movement vector can be acquired by connecting the vehicle positions between consecutive frames. The movement direction vector can be acquired by connecting the vehicle position in the frame in which the target vehicle was first detected (i.e., the position where the target vehicle appeared in the camera image) with the vehicle position in the frame in which the target vehicle was last detected (i.e., the position where the target vehicle disappeared from the camera image).

[0028] The vector conversion unit 62 performs vector conversion processing on the vehicle direction vector obtained by the vector acquisition unit 61 to compensate for changes in the movement direction vector due to the PTZ. Specifically, the vector conversion unit 62 converts the movement direction vector into a coordinate system of the initial setting angle of view using an affine transformation matrix or the like, using information related to changes in the angle of view of the camera 10 (internal parameters related to the orientation, focal length, etc. of the camera 10). Naturally, if the angle of view of the camera 10 has not changed from the initial setting angle of view, the conversion by the vector conversion unit 62 becomes an identity transformation. Note that if information related to changes in the angle of view cannot be acquired, the road extension direction is reset in the road direction holding unit 63, and the system operation is resumed.

[0029] The directional similarity calculation unit 64 calculates the similarity between the movement direction vector corrected by the vector conversion unit 62 and the road direction vector stored in the road direction storage unit 63. In this example, cosine similarity, which considers only the direction of each vector without considering the magnitude of each vector, is used as the similarity between the movement direction vector and the road direction vector, but other methods may be used to calculate the similarity. The cosine similarity is an index that represents the similarity in the direction of vectors in a range from -1 to 1, with the closer the cosine similarity is to 1, the closer the vectors are to the directions of the other vectors, and the closer the cosine similarity is to -1, the closer the vectors are to the directions of the other vectors. Therefore, a vehicle whose output from the directional similarity calculation unit 64 is close to 1 is likely to be moving in the forward direction of the road direction vector, and a vehicle whose output from the directional similarity calculation unit 64 is close to -1 is likely to be moving in the reverse direction of the road direction vector.

[0030] The direction determination unit 65 determines the direction of vehicle movement by threshold processing of the similarity obtained by the directional similarity calculation unit 64, and counts the number of vehicles. Specifically, when the absolute value of the similarity obtained by the directional similarity calculation unit 64 exceeds a predetermined threshold, the direction determination unit 65 counts the vehicle as having traveled in the direction of the road direction vector. It is desirable to use a value corresponding to the shape of the road portion to be monitored as the threshold used in the threshold processing. Specifically, when the road portion to be monitored is straight, a value close to 1 but equal to or greater than 0.7 is set. Furthermore, when the road portion to be monitored is curved, a small value may be set when the radius R of the curve is small, and a value of approximately 0.3 is set when R is particularly small.

[0031] If the road is bidirectional, the direction determination unit 65 counts the number of vehicles in the uphill and downhill directions separately. Here, if only one road direction vector is set for one road, one for the uphill and one for the downhill directions, the number of vehicles traveling in the direction corresponding to the road direction vector for which a positive similarity is obtained may be counted. Also, if only one road direction vector is set for one road, the direction in which the vehicles are traveling, either the forward or reverse direction of the road direction vector, may be determined based on the positive or negative similarity, and the number of vehicles may be counted for each direction. That is, if the similarity is positive, the number of vehicles traveling in the forward direction of the road direction vector may be counted, and if the similarity is negative, the number of vehicles traveling in the reverse direction of the road direction vector may be counted. The results of the counting of the number of vehicles by direction by the direction determination unit 65 are stored in the count recording unit 30.

[0032] In the conventional system, the accuracy of counting depends on the accuracy of the trajectory of the vehicle crossing the virtual line, whereas in the present system, it depends on how to reduce errors in vehicle IDs that occur during vehicle tracking, i.e., ID SW (ID switch), in which the same vehicle ID is assigned to different vehicles. To overcome this problem of IDSW, the present system is designed to detect the occurrence of IDSW and compensate for it using the tracking result correction unit 53.

[0033] 6 shows an example of a flowchart regarding compensation of IDSW by the tracking result compensation unit 53. The tracking result compensation unit 53 performs the process shown in FIG. 6 to compensate for the case where the same vehicle ID is assigned to different vehicles.

[0034] First, the tracking result correction unit 53 acquires an inter-frame motion vector Vf representing the motion direction of the vehicle between frames, with the previous frame as the start point and the current frame as the end point, for each vehicle ID obtained by the vehicle tracking process in the vehicle tracking unit 51 (step S11). Next, for each inter-frame motion vector Vf, the tracking result correction unit 53 calculates the similarity with the road direction vector stored in the road direction storage unit 63 in the same manner as the direction similarity calculation unit 64 (step S12).

[0035] Next, threshold processing is performed on the similarity calculated for each inter-frame motion vector Vf. Specifically, it is determined whether the absolute value of the similarity between the inter-frame motion vector Vf and the road direction vector is smaller than the aforementioned threshold, or whether the sign of the inter-frame motion vector Vf has reversed relative to the previous frame (step S13). Here, a similarity close to 0 indicates that the vehicle has moved in a direction perpendicular to the road direction, and a reversal of the sign indicates that the vehicle has suddenly moved in the opposite direction. Since a vehicle typically does not suddenly move in a direction perpendicular to the road direction or in the opposite direction, it can be assumed that the vehicle ID is assigned to multiple vehicles. Note that the tracking result correction unit 53 in this example uses the same threshold as that used by the direction determination unit 65, but a different threshold may also be used.

[0036] If the tracking result correction unit 53 finds an inter-frame movement vector Vf that satisfies the above conditions (step S13; Yes), it discards the corresponding vehicle ID and its trajectory data (step S14). That is, the tracking result correction unit 53 deletes the vehicle ID and its trajectory data related to the inter-frame movement vector Vf that satisfies the above conditions from the tracking results obtained by the vehicle tracking unit 51. By performing the above processing, the movement direction vector obtained for the vehicle ID in which IDSW occurred is excluded from the counting target, while the movement direction vectors of the remaining vehicle IDs represent the correct movement direction of each vehicle, so it is expected that the counting will be accurate.

[0037] 7A and 7B, the difference in vehicle ID assignment depending on whether or not the tracking result correction unit 53 corrects the tracking result will be described. FIG. 7A shows an example of a vehicle ID before the tracking result correction is performed. FIG. 7B shows an example of a vehicle ID after the tracking result correction is performed. As shown in FIG. 7A, due to the occurrence of an IDSW, the same vehicle ID = 1 is assigned to vehicle 1 traveling in the uphill direction and vehicle 2 traveling in the downhill direction. Furthermore, vehicle ID = 2 is assigned to vehicle 2 before the IDSW occurs, and vehicle ID = 3 is assigned to vehicle 1 after the IDSW occurs.

[0038] If tracking result correction is not performed, as shown in FIG. 7A , the movement direction vector calculated from the start and end positions of the trajectory of vehicle ID = 1 is close to being perpendicular to the road direction vector, and the similarity between these vectors is close to 0. Therefore, the similarity between the movement direction vector of vehicle ID = 1 and the road direction vector is below the threshold, and there is a possibility that the number of vehicles will not be counted correctly. On the other hand, by performing tracking result correction, as shown in FIG. 7B , the data for vehicle ID = 1 is discarded, but the uphill movement direction vector for vehicle 1 is obtained from the data for vehicle ID = 3, and the downhill movement direction vector for vehicle 2 is obtained from the data for vehicle ID = 2. As a result, it is possible to count one vehicle each in the uphill and downhill directions, thereby obtaining accurate count results.

[0039] As described above, in the traffic volume counting system of this example, the vehicle tracking unit 51 analyzes the image captured by the camera 10 to acquire the vehicle trajectory, the vector acquisition unit 61 acquires the movement direction vector indicating the direction of the vehicle trajectory from the start point to the end point, the direction similarity calculation unit 64 calculates the similarity between the movement direction vector and a road direction vector preset with respect to the reference angle of view of the camera 10, and the direction determination unit 65 determines that a vehicle is traveling in the direction along the road direction vector when the absolute value of the similarity is greater than a predetermined threshold, and counts the number of vehicles. In other words, instead of determining the intersection of the vehicle trajectory and the virtual line as in the conventional method, by determining the similarity between the road direction vector and the movement direction vector, vehicles traveling on the road can be appropriately detected, thereby enabling the number of vehicles to be counted with high accuracy.

[0040] The traffic volume count system of this example also has a vector conversion unit 62 that performs vector conversion processing on the movement direction vector obtained by the vector conversion unit 61 in response to changes in the angle of view of the camera 10, and is configured to calculate the similarity between the movement direction vector after vector conversion by the vector conversion unit 62 and a road direction vector. Therefore, even if the angle of view changes due to a change in the orientation or zoom of the camera 10, there is no need to reset the road direction vector. Note that instead of performing vector conversion processing on the movement direction vector, a configuration may be adopted in which vector conversion processing is performed on the road direction vector.

[0041] The traffic volume counting system of this example also includes a tracking result correction unit 53 that acquires inter-frame motion vectors that represent the direction of vehicle movement between frames based on the vehicle trajectories obtained by the vehicle tracking unit 51, and excludes the vehicle trajectory related to the inter-frame motion vector from counting if the absolute value of the similarity between the road direction vector and the inter-frame motion vector is smaller than a predetermined threshold, or if the sign of the inter-frame motion vector is reversed compared to the immediately preceding frame. This reduces the impact of IDSW occurrence on vehicle counting, enabling more accurate counting.

[0042] Here, if it is not possible to obtain information relating to changes in the angle of view of the camera 10 (internal parameters relating to the orientation and focal length of the camera 10, etc.) due to the specifications of the camera 10, it is necessary to manually reset the road direction vector in the road direction storage unit 63. However, just as with the initial setting, the setting of the road direction vector can be changed simply by the operator specifying the start and end points of the road on the screen, and this is not a task that requires an expert, so it remains more convenient than conventional systems.

[0043] Although the embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. The present invention can take on various other embodiments, and various modifications such as omissions and substitutions can be made without departing from the spirit of the present invention. These embodiments and modifications thereof are included in the scope and spirit of the invention described in this specification, etc., and are included in the invention described in the claims and their equivalents.

[0044] Furthermore, the present invention can be provided not only as devices such as those described above or as systems composed of these devices, but also as methods executed by these devices, programs for realizing the functions of these devices using a processor, and storage media for storing such programs in a computer-readable manner.

[0045] The present invention can be used in a traffic counting method and a traffic counting system for counting the number of vehicles traveling on a road.

[0046] 10: Camera, 20: Transmission unit, 30: Count recording unit, 40: Traffic volume count processing unit, 41: Vehicle tracking unit, 42: Trajectory storage unit, 43: Virtual line storage unit, 44: Intersection determination unit, 50: Traffic volume count processing unit, 51: Vehicle tracking unit, 52: Movement direction determination unit, 53: Tracking result correction unit, 61: Vector acquisition unit, 62: Vector conversion unit, 63: Road direction storage unit, 64: Direction similarity calculation unit, 65: Direction determination unit

Claims

1. A traffic volume counting method for counting the number of vehicles based on images captured by a camera of a road, comprising: a process for analyzing the image captured by the camera to obtain a vehicle trajectory; a process for obtaining a movement direction vector indicating the direction of an end point position relative to a start point position of the vehicle trajectory; a process for calculating a similarity between a road direction vector set in advance for a reference angle of view of the camera and the movement direction vector; and a process for determining that the vehicle is traveling in a direction along the road direction vector and counting the number of vehicles when the absolute value of the similarity is greater than a predetermined threshold value.

2. A traffic volume counting method according to claim 1, characterized in that when only one road direction vector is set for one road, the process of counting the number of vehicles determines whether the vehicle is traveling in the forward or reverse direction of the road direction vector depending on whether the similarity is positive or negative, and counts the number of vehicles by direction.

3. A traffic volume counting method according to claim 1, characterized in that an inter-frame movement vector representing the direction of movement of the vehicle between frames is obtained based on the vehicle's trajectory, and if the absolute value of the similarity between the road direction vector and the inter-frame movement vector is smaller than the threshold value, the vehicle's trajectory related to the inter-frame movement vector is excluded from being counted.

4. A traffic volume counting method according to claim 1, characterized in that an inter-frame movement vector representing the direction of movement of the vehicle between frames is obtained based on the vehicle's trajectory, and if the sign of the inter-frame movement vector is reversed with respect to the immediately preceding frame, the vehicle's trajectory relating to the inter-frame movement vector is excluded from the counting target.

5. A traffic volume counting method according to claim 1, characterized in that the road direction vector or the movement direction vector is subjected to a vector conversion process corresponding to a change in the angle of view of the camera, and then the similarity is calculated.

6. A traffic volume counting system that counts the number of vehicles based on images captured by a camera of a road, comprising: a vehicle tracking unit that analyzes the image captured by the camera to obtain a vehicle trajectory; a vector acquisition unit that acquires a movement direction vector indicating the direction of an end point relative to a start point related to the vehicle trajectory; a direction similarity calculation unit that calculates a similarity between a road direction vector that is set in advance with respect to a reference angle of view of the camera and the movement direction vector; and a direction determination unit that, when the absolute value of the similarity is greater than a predetermined threshold, determines that the vehicle is traveling in a direction along the road direction vector and counts the number of vehicles.

Images