JP2006202047A - Moving-object detection device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving-object detection device which can very precisely trace a moving object, and also provide a moving-object detection method. <P>SOLUTION: This is a moving-object detection device 1 which detects a moving object from a picked-up image. In addition, the device comprises; an image pickup means 2; a moving-object detection means 13 which detects a moving object from an picked-up image taken with the image pickup means 2; a direction detection means 15 which detects the direction of the picked-up image detected by the moving-object detection means 13; a movement detection means 16 which detects the movement of the moving object from the positions of moving objects detected from each of the picked-up images taken in different time by the image pickup means 2; and an area forecasting means 17 which forecasts an area, in which the moving object in the picked-up image exists, based on the direction of the moving object detected by the direction detection means 15 and the movement of the moving object detected by the movement detection means 16. The moving object is detected from the inside of the area forecasted by the area forecasting means 17. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a moving object detection apparatus and a moving object detection method for detecting a moving object from a captured image and tracking the detected moving object.

In recent years, an apparatus for detecting a pedestrian existing in front of a vehicle has been proposed for the purpose of driving assistance and safety improvement (see Patent Document 1). In the pedestrian detection device described in Patent Document 1, the center point of the pedestrian's center line is sequentially obtained from each captured image captured at regular intervals, and the pedestrian's center point is determined from each center point between successive captured images. A movement vector is obtained, and a pedestrian is tracked using the movement vector. This movement vector indicates the direction and size of the direction in which the pedestrian has moved in the past, and is used for prediction of the direction in which the pedestrian moves.
JP 2004-86417 A

  In the pedestrian tracking based on the conventional movement vector, the future movement direction is predicted using only the movement direction of the past pedestrian. However, pedestrians are not only moving in a certain direction, but may suddenly change the direction of movement, such as when crossing a pedestrian crossing from a sidewalk. When the pedestrian moves in a direction different from the past as described above, the movement vector obtained from the past information does not include information for changing the moving direction. For this reason, if a pedestrian is tracked with a future captured image based on this movement vector, the pedestrian moves away from the direction of the movement vector, so there is a high possibility that tracking will fail. In particular, when the movement of the pedestrian is fast, the amount of movement of the pedestrian between successive captured images increases, so the probability of failure increases.

  Therefore, an object of the present invention is to provide a moving object detection apparatus and a moving object detection method that can track a moving object with high accuracy.

  A moving object detection device according to the present invention includes an imaging unit, a moving object detection unit that detects a moving object from an image captured by the imaging unit, and a direction detection unit that detects the direction of the moving object detected by the moving object detection unit. A motion detection means for detecting the movement of the moving object from the position of the moving object detected from each captured image captured at different times by the imaging means, and a direction and motion detection means for the moving object detected by the direction detection means. An area predicting unit that predicts an area where the moving object exists in the captured image based on the detected movement of the moving object, and detects the moving object from the area predicted by the area predicting unit.

  In this moving object detection device, an image is picked up by an image pickup means, and a moving object (for example, a pedestrian, bicycle, animal) is detected from the picked-up image by the moving object detection means. Then, in the moving object detection means, the direction detection means detects the direction in which the moving object is facing from the current captured image. From the direction of the moving object, the direction in which the moving object is going at the present time is known, and the direction in which the moving object is going to be moved can be estimated. The moving object detection unit detects the movement of the moving object from the position of the moving object detected from each captured image at different times before the current time by the movement detection unit. By the movement of the moving object, the direction in which the moving object has moved up to the present time can be determined. Therefore, when the direction and movement of the moving object are in the same direction, it can be estimated that the moving object is going to move in the same direction as in the past after the current time. On the other hand, when the direction and movement of the moving object are clearly different directions, it can be estimated that the moving object is moving in a different direction from the past in the future. Therefore, in the moving object detection device, an area in which a moving object is present in a future captured image is predicted by the area predicting unit based on the direction and movement of the moving object. Then, the moving object detection device tracks the already detected moving object by detecting the moving object from the predicted region. As described above, the moving object detection apparatus can predict the region where the moving object exists in the future captured image with high accuracy even when the moving object is moving in a certain direction or when the moving direction is changed. it can. As a result, it is possible to track a moving object with high accuracy.

  The moving object detection method according to the present invention includes an imaging step, a moving object detection step that detects a moving object from the captured image captured in the imaging step, and a direction detection step that detects the direction of the moving object detected in the moving object detection step. And a motion detection step for detecting the motion of the moving object from the position of the moving object detected from each captured image captured at different times in the imaging step, and a direction and motion detection step of the moving object detected in the orientation detection step. A region prediction step of predicting a region where the moving object is present in the captured image based on the detected movement of the moving object, and detecting the moving object from within the region predicted in the region prediction step. This moving object detection method has the same effects as the moving object detection apparatus.

  According to the present invention, a moving object can be tracked with high accuracy.

  Hereinafter, embodiments of a moving object detection device and a moving object detection method according to the present invention will be described with reference to the drawings.

  In the present embodiment, the present invention is applied to a pedestrian detection device mounted on a vehicle. The pedestrian detection device according to the present embodiment detects a pedestrian existing in front of the vehicle and provides the driver with information on the detected pedestrian. In particular, in the pedestrian detection device according to the present embodiment, for a pedestrian that has been detected once, an area existing in the next frame is predicted, and the pedestrian is traced using the predicted area.

  With reference to FIGS. 1-3, the pedestrian detection apparatus 1 is demonstrated. FIG. 1 is a configuration diagram of a pedestrian detection device according to the present embodiment. FIG. 2 is an image showing a basic flow of the tracking process in the pedestrian detection apparatus of FIG. 1, (a) is an image in which a new pedestrian is detected in a certain frame, and (b) is in the next frame. It is an image showing a prediction area where a tracking pedestrian exists, (c) is an image in which a tracking pedestrian is detected in the next frame, and (d) is a tracking pedestrian in the next frame (E) is an image in which a tracking pedestrian is detected in the next next frame, and (f) is a prediction in which there is a tracking pedestrian in the next next frame. It is the image which detected the area | region and the pedestrian in tracking. FIG. 3 is an explanatory diagram of a method of calculating the direction and size of a pedestrian's movement.

  The pedestrian detection device 1 images the front of the vehicle and detects a pedestrian from the captured image. At this time, the pedestrian detection device 1 detects a pedestrian from the entire captured image, and sequentially tracks a newly detected pedestrian as a tracking target while predicting a region where the pedestrian being tracked exists, while tracking from the predicted region. The pedestrian is tracked by detecting the pedestrian. In particular, in the pedestrian detection apparatus 1, in order to track a pedestrian with high accuracy, information on the direction of the pedestrian and the movement of the pedestrian is used when the region is predicted. For this purpose, the pedestrian detection apparatus 1 includes a camera 2, a monitor 3, and an image ECU [Electronic Control Unit] 4. The image ECU 4 includes a detection result storage memory 10, a prediction area storage memory 11, an image input unit 12, a walking. A person detection unit 13, a tracking detection unit 14, a direction calculation unit 15, a motion calculation unit 16, a region prediction unit 17, and a result display unit 18 are configured.

  In the present embodiment, the camera 2 corresponds to the imaging unit described in the claims, the pedestrian detection unit 13 corresponds to the moving object detection unit described in the claims, and the direction calculation unit 15 includes It corresponds to the direction detection means described in the claims, the motion calculation unit 16 corresponds to the motion detection means described in the claims, and the region prediction unit 17 corresponds to the region prediction means described in the claims. To do.

  The camera 2 is attached in front of a vehicle on which the pedestrian detection device 1 is mounted, and images the front of the vehicle. This captured image may be a color image or a black and white image. The camera 2 transmits the captured image data to the image ECU 4. The camera 2 has a wide imaging range in the left-right direction, and can capture images such as sidewalks and roadside zones provided outside the roadway.

  The monitor 3 is attached to the front of the passenger compartment at a position where the driver can visually recognize it. In the monitor 3, an image signal is transmitted from the image ECU 4, and an image corresponding to the image signal is displayed on the screen.

  The image ECU 4 includes a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like, and each processing unit and each memory are configured. The image ECU 4 takes captured image data from the camera 2, detects a pedestrian in front of the vehicle from the captured image, and displays the detected pedestrian information on the monitor 3.

  The detection result storage memory 10 is a memory configured to store the information about the pedestrian detected by the pedestrian detection unit 13 and the information about the pedestrian being tracked detected by the tracking detection unit 14. The information regarding the pedestrian detected by the pedestrian detection unit 13 is information on the pedestrian detected in the current frame, and is a position and a region in the image of the pedestrian. The information on the pedestrian being tracked is information on the pedestrian currently being tracked, and is the position, area, orientation, and movement (direction and size) in the image of the pedestrian being tracked.

  The prediction area storage memory 11 is a memory configured to store a prediction area obtained by the area prediction unit 17 in the RAM. The predicted region is a region predicted as a region where a pedestrian currently being tracked is likely to exist in the next frame.

  The image input unit 12 takes captured image data from the camera 2 and stores the captured image data in a predetermined storage area (RAM).

  Each time captured image data is input from the camera 2, the pedestrian detection unit 13 detects a pedestrian by pattern matching using a template from the entire captured image of the current frame. At this time, the detected pedestrian is cut out in a rectangular area WA including the entire pedestrian corresponding to the template image (rectangular area image) (see FIG. 2A). And in the pedestrian detection part 13, the center position of the cut-out rectangular area WA is made into the pedestrian position WP (refer Fig.2 (a)). Further, when the pedestrian detection unit 13 can detect pedestrians from the entire captured image, the positions WP (coordinates on the image) and the rectangular areas WA (coordinates on the images of the four end portions) of all detected pedestrians. ) Is stored in the detection result storage memory 10. The detected pedestrian includes a pedestrian (new pedestrian) NW newly detected from the current frame (see FIG. 2A) and a pedestrian tracked from a past frame (pedestrian being tracked). ) CW (see FIGS. 2C, 2E, and 2F).

  In the tracking detection unit 14, when there is a tracking pedestrian (when the information of the tracking pedestrian is stored in the detection result storage memory 10), the corresponding prediction region of the tracking pedestrian from the prediction region storage memory 11. Extract PA. Then, the tracking detection unit 14 cuts out the image of the extracted prediction area PA from the captured image, and detects the pedestrian CW being tracked from the image of the prediction area PA by pattern matching using a template (FIG. 2 (c), ( e) and (f)). At this time, the detected tracking pedestrian CW is cut out in a rectangular area WA including the entire pedestrian corresponding to the template image (rectangular area image) (see FIG. 2C, etc.). Then, in the tracking detection unit 14, when the tracking pedestrian can be detected from the prediction area PA, the center position of the cut out rectangular area WA is set as the position WP of the tracking pedestrian (see FIG. 2 (c), etc.) The detected position WP and the rectangular area WA of the tracking pedestrian are stored in the detection result storage memory 10 as information of the tracking pedestrian. On the other hand, when the tracking pedestrian cannot be detected from the prediction area PA, the tracking detection unit 14 deletes the information of the tracking pedestrian from the prediction area storage memory 11. Incidentally, the tracking detection unit 14 performs the above-described processing for all the tracking pedestrians stored in the prediction area storage memory 11.

  Further, in the tracking detection unit 14, a prediction region is included in the pedestrian detected from the captured image of the current frame (that is, in the pedestrian detected by the pedestrian detection unit 13 stored in the detection result storage memory 10). When the same pedestrian as the tracking pedestrian detected from the PA exists, the information on the same pedestrian is deleted from the detection result storage memory 10. Therefore, when the tracking detection unit 14 finishes tracking all pedestrians being tracked, the detection result storage memory 10 stores new information newly detected from the current frame as pedestrian information detected by the pedestrian detection unit 13. Only pedestrian information is stored.

  The direction calculation unit 15 calculates the direction FD that is facing in the current frame of the new pedestrian and the tracking pedestrian (see FIGS. 2B, 2C, and 2E). Then, the direction calculation unit 15 stores the calculated direction FD information in the detection result storage memory 10 in association with the pedestrian. Since the information on the direction is obtained from the image of the current frame, the direction in which the pedestrian is heading at the present time can be known regardless of the movement of the pedestrian in the past.

  As a method for calculating the direction, a matching is performed between the detected image of the rectangular area WA of the pedestrian and the template image of the pedestrian having a different direction, and the pedestrian is determined from the direction in which the template having the highest degree of correlation is directed. The direction FD is obtained. This template is a template for pedestrians whose directions are changed at predetermined angles over 360 °. The processing load is reduced as the predetermined angle is increased, and the accuracy is improved as the predetermined angle is decreased. When the degree of correlation between two templates at adjacent angles is high, the direction of the pedestrian is obtained by interpolation from the direction in which the two templates are facing. The method for obtaining this direction is described in, for example, H. Shimizu and T. Poggio, “Direction Estimation of Pedestrian from Multiple Still Images”, IEEE Intelligent Vehicles Symposium, 2004, Parma, Italy.

  The motion calculation unit 16 calculates the direction MD and the magnitude of the pedestrian's movement from each position WP of the tracked pedestrian's current frame and the previous frame (see FIGS. 2C and 2E). Then, the motion calculation unit 16 stores the calculated motion direction MD and size information in the detection result storage memory 10 in association with the tracking pedestrian. Since this movement information is obtained from the information of the past frame including the current frame, the direction and amount of movement of the pedestrian in the past can be known. In addition, since this motion information is obtained from information of a plurality of frames, highly accurate information can be obtained.

  As a motion calculation method, a vector starting from the position WP1 of the tracking pedestrian CW1 in the current frame as an end point and starting from the position WP2 of the tracking pedestrian CW2 in the previous frame is obtained, and the angle of the vector is determined as the motion angle. The direction is MD, and the length of the vector is the magnitude of motion (see FIG. 3).

  The area predicting unit 17 predicts an area PA where the pedestrian in tracking exists in the next frame from the direction FD of the pedestrian and the direction MD and size of the movement of the pedestrian in tracking (FIG. 2B, FIG. 2D). ) And (f)). Then, the region prediction unit 17 stores the predicted region PA predicted in association with the tracking pedestrian in the prediction region storage memory 11. The prediction method for this area will be described separately for a new pedestrian and a tracking pedestrian.

  The case of a new pedestrian will be described. In the case of a new pedestrian NW, there is no movement information because it is a pedestrian newly detected from the current frame. Therefore, the prediction area PA is obtained using only the direction FD. As the prediction area PA, a rectangular area WA of the pedestrian NW is included at the end, and the horizontal length and the vertical length of the rectangular area WA are each extended by a predetermined amount along the direction FD. (See FIG. 2 (a) → (b)). The predetermined magnification is, for example, about 1.5 times in the vertical direction and about twice in the horizontal direction in consideration of the size of the rectangular area with respect to the entire image, the time interval between successive frames, the maximum speed of the pedestrian, and the like. This predetermined multiple may be a fixed value or may be a variable value in consideration of the position of the pedestrian in the depth direction. When the orientation FD is horizontal in the image, the prediction area PA is obtained by extending only the horizontal length by a predetermined multiple. When the orientation FD is vertical in the image, only the vertical length is multiplied by a predetermined multiple. The prediction area PA is extended to.

  The case of a pedestrian during tracking will be described. When the pedestrian is moving in a certain direction, the direction FD and the direction MD of movement match or substantially match (see FIG. 2C). In this case, the prediction area PA is obtained using the information on the movement of the pedestrian using the information of a plurality of frames. On the other hand, when the moving direction is changed, for example, when a pedestrian walking on the sidewalk crosses a pedestrian crossing, the direction FD and the direction MD of movement are clearly different (see FIG. 2E). In this case, since past movement information cannot be used, the prediction area PA is obtained using the pedestrian orientation FD.

  Therefore, first, an absolute value of a difference between the direction FD and the motion direction MD is obtained, and it is determined whether or not the absolute value is larger than a threshold value. When the value is equal to or smaller than the threshold value, the prediction area PA is obtained using the information on the movement of the pedestrian, and when larger than the threshold value, the prediction area PA is obtained using the direction FD of the pedestrian. The threshold is an angle for determining whether or not the direction in which the pedestrian is moving has been changed, and is set in consideration of the accuracy of the direction FD and the direction MD of the movement, and is, for example, 45 °. In the case of the threshold value or less, the prediction area PA includes the rectangular area WA of the pedestrian CW being tracked at the end, and the horizontal length and the vertical length of the rectangular area WA along the direction of the motion MD. It is set as the area | region extended by the magnitude | size of the movement (refer FIG.2 (c)-> (d)). On the other hand, when larger than the threshold value, the prediction area PA includes the rectangular area WA of the pedestrian CW being tracked at the end, and the horizontal length and the vertical length of the rectangular area WA along the direction FD. It is assumed that each of the areas is a predetermined multiple (see FIG. 2 (e) → (f)).

  In the result display unit 18, when a pedestrian exists in the captured image of the current frame (that is, when information on a new pedestrian or / and information on a pedestrian being tracked is stored in the detection result storage memory 10), Data of an image obtained by adding a line indicating the rectangular area WA of each pedestrian to the entire captured image of the current frame is transmitted to the monitor 3 as an image signal. At this time, a line indicating a rectangular area WA only for a pedestrian within a predetermined distance from the vehicle may be added, or a line indicating a rectangular area WA only for a pedestrian existing on a roadway such as on a pedestrian crossing may be used. It may be an added image. Alternatively, the entire captured image of the current frame may be eliminated and an image showing only pedestrians may be used.

  With reference to FIG. 1, operation | movement of the pedestrian detection apparatus 1 is demonstrated. In particular, the processing in the image ECU 4 will be described along the flowchart shown in FIG. FIG. 4 is a flowchart showing a flow of processing in the image ECU of the pedestrian detection apparatus of FIG.

  The camera 2 captures the front of the host vehicle at regular time intervals and transmits data of the captured image to the image ECU 4. Then, the image ECU 4 takes in the data of the captured image at regular intervals and stores it in a predetermined memory area (S1).

  The image ECU 4 detects a pedestrian existing in the captured image of the current frame by pattern matching. If the pedestrian can be detected, the position and the rectangular area of the pedestrian are stored in the detection result storage memory 10 (S2). ). Then, the image ECU 4 determines whether or not a pedestrian has been detected from the captured image of the current frame (S3).

  When it is determined in S3 that the pedestrian has been detected, the image ECU 4 includes a pedestrian that matches the pedestrian being tracked stored in the detection result storage memory 10 among the pedestrians detected in the captured image of the current frame. It is determined whether or not it exists (S4). If it is determined in S4 that there is no pedestrian that matches the pedestrian being tracked, all the pedestrians detected in the current frame are new pedestrians detected for the first time in the current frame, so the image ECU 4 proceeds to the processing of S11. Then, the orientations of all the new pedestrians are calculated by template matching, and the orientation information is associated with each pedestrian and stored in the detection result storage memory 10 (S11).

  On the other hand, if it is determined in S3 that the pedestrian cannot be detected, the image ECU 4 determines whether or not there is a tracking pedestrian stored in the detection result storage memory 10, and the stored tracking is in progress. If there are pedestrians, it is determined whether or not the tracking process has been completed for all of the pedestrians being tracked (S5).

  If it is determined in S4 that there is a pedestrian that matches the pedestrian being tracked, or if it is determined in S5 that processing has not been completed for all the pedestrians being tracked, the image ECU 4 stores the detected result in the detection result storage memory 10. The stored tracking pedestrians are selected one by one, and the predicted region of the selected tracking pedestrian is extracted from the prediction region storage memory 11. Then, the image ECU 4 cuts out the extracted prediction area from the captured image of the current frame. Further, the image ECU 4 detects the pedestrian being tracked from the image of the predicted region by pattern matching, and if the pedestrian being tracked can be detected, the position and rectangular area of the pedestrian being tracked are detected result storage memory 10. (S6). Then, the image ECU 4 determines whether or not the tracking pedestrian has been detected from the image of the prediction region (S7).

  If it is determined in S7 that the tracking pedestrian cannot be detected, it can be assumed that the pedestrian who has been tracking up to the previous frame has gone out of the imaging range of the camera 2 in the current frame. Is deleted from the detection result storage memory 10 and the prediction area storage memory 11. Further, the image ECU 4 determines whether or not there is a new pedestrian detected for the first time in the current frame among the pedestrians detected in the current frame (S8). If it is determined in S8 that a new pedestrian exists, the image ECU 4 shifts to the processing in S11, calculates the orientations of all the new pedestrians by matching with the template, and handles the direction information for the pedestrians. At the same time, it is stored in the detection result storage memory 10 (S11). On the other hand, when it is determined in S8 that there is no new pedestrian, the image ECU 4 returns to the process of S5, and whether or not the process has been completed for all the tracking pedestrians stored in the detection result storage memory 10. Is determined (S5).

  If it is determined in S7 that the pedestrian being tracked can be detected, the image ECU 4 determines whether or not the same pedestrian as the detected pedestrian exists in the pedestrians detected in the current frame. (S9). When it is determined in S9 that the same pedestrian exists, the image ECU 4 determines the same pedestrian information from the pedestrian information detected in the current frame stored in the detection result storage memory 10. Is deleted (S10). Therefore, in the detection result storage memory 10, finally, only information on a new pedestrian detected for the first time in the current frame remains as information on the pedestrian detected in the current frame.

  When it is determined in S9 that the same pedestrian exists, and when the process of S10 ends, the image ECU 4 calculates the detected direction of the pedestrian being tracked by matching with the template, and uses the information on the direction as the pedestrian. And stored in the detection result storage memory 10 (S11). With this direction information, the direction in which the pedestrian is about to move can be predicted. Further, the image ECU 4 calculates the direction and size of the movement of the tracking pedestrian from the difference between the detected position of the pedestrian in the previous frame and the position in the current frame, and is tracking the movement information. Corresponding to the pedestrian, it is stored in the detection result storage memory 10 (S12). From this movement information, the direction and amount of movement of the tracking pedestrian so far can be known.

  Further, the image ECU 4 calculates a prediction area in the captured image of the next frame from the orientation information in the case of a new pedestrian for the detected pedestrian, and the orientation information and A predicted area in the captured image of the next frame is calculated from the motion information, and the predicted area is stored in the predicted area storage memory 11 in association with the pedestrian being tracked (S13). At this time, for the new pedestrian, only the information on the direction in the current frame is used, and therefore the region to be moved can be predicted from the information on the direction. On the other hand, for a tracking pedestrian, if the direction matches the direction of the movement, the area to be moved can be predicted from the reliable movement information, and the direction and the direction of the movement If they are different, the area to be moved can be predicted from the direction information.

  When the region is predicted, the image ECU 4 returns to the process of S5, and determines whether or not the process has been completed for all the tracking pedestrians stored in the detection result storage memory 10 (S5). When it is determined in S5 that the processing has been completed for all the pedestrians being tracked, the image ECU 4 displays the entire captured image of the current frame when a new pedestrian and / or a pedestrian being tracked is detected in the current frame. Data of an image to which a line indicating the rectangular area of each pedestrian is added is transmitted as an image signal to the monitor 3 (S14). When this image signal is received, the monitor 3 displays an image in which the area where the pedestrian is present is emphasized in the captured image of the current frame. Further, in the image ECU 4, when new pedestrian information exists in the detection result storage memory 10, the new pedestrian information is newly added and stored as information on the pedestrian being tracked in the detection result storage memory 10. , Delete that new pedestrian information. When all the processes in the current frame are completed, the image ECU 4 returns to the process of S1 and acquires the captured image of the next frame (S1).

  According to this pedestrian detection device 1, by using not only the past pedestrian movement but also the current pedestrian direction, the region where the pedestrian being tracked is present in the next frame is predicted with high accuracy. And tracking performance is improved. In particular, in the pedestrian detection device 1, even when the pedestrian suddenly changes the direction of movement, the pedestrian detection device 1 uses a direction in which the direction of movement of the pedestrian from the current time can be estimated, so that tracking is possible.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, in this embodiment, the present invention is applied to a pedestrian detection device that detects a pedestrian mounted on a vehicle, but may be applied to a device that detects other moving objects such as bicycles and animals other than pedestrians. It may be applied to a configuration in which a moving object detection device is provided on the infrastructure side and information detected on the infrastructure side is transmitted to the vehicle side, or to a robot having a function of tracking a moving object such as a human. May be.

  Also, in this embodiment, the detected pedestrian information is displayed on the monitor. However, when lanes (white lines), roadways and sidewalks, and pedestrian crossings are also detected, such information is also displayed on the monitor. Further, it may be configured to detect whether or not a pedestrian has entered the lane and output an alarm or the like when the pedestrian has entered, or may be configured to walk without displaying on the monitor. The presence of the person may be notified by voice or buzzer.

  Further, in the present embodiment, each processing unit is configured by software formed by executing each program by the CPU, but each processing unit may be configured by hardware.

  In the present embodiment, the pedestrian orientation is obtained by matching with the template, and the pedestrian movement is obtained from the position of the pedestrian between the frames. However, the pedestrian orientation and the pedestrian movement are obtained. Is not limited to these, and other methods may be used.

  In this embodiment, the movement is obtained from the position of each pedestrian in the current frame and the previous frame. However, the movement may be obtained by using the positions of pedestrians in the previous frame before the previous frame. Good.

It is a block diagram of the pedestrian detection apparatus which concerns on this Embodiment. It is an image which shows the basic flow of the tracking process in the pedestrian detection apparatus of FIG. 1, (a) is an image in which a new pedestrian is detected in a certain frame, and (b) is a pedestrian in tracking in the next frame. (C) is an image in which a tracking pedestrian is detected in the next frame, and (d) is a prediction region in which the tracking pedestrian exists in the next frame. (E) is an image in which a tracking pedestrian is detected in the next next frame, and (f) is a tracking region in which the tracking pedestrian exists in the next next frame and the tracking It is the image which detected the pedestrian. It is explanatory drawing of the calculation method of the direction and magnitude | size of a pedestrian's movement. It is a flowchart which shows the flow of a process in image ECU of the pedestrian detection apparatus of FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Pedestrian detection apparatus, 2 ... Camera, 3 ... Monitor, 4 ... Image ECU, 10 ... Detection result storage memory, 11 ... Prediction area | region storage memory, 12 ... Image input part, 13 ... Pedestrian detection part, 14 ... Tracking Detection unit, 15 ... orientation calculation unit, 16 ... motion calculation unit, 17 ... region prediction unit, 18 ... result display unit

Claims (2)

Imaging means;
Moving object detection means for detecting a moving object from a captured image captured by the imaging means;
Direction detecting means for detecting the direction of the moving object detected by the moving object detecting means;
Motion detection means for detecting the movement of the moving object from the position of the moving object detected from each captured image captured at different times by the imaging means;
A region prediction unit that predicts a region where a moving object exists in a captured image based on the direction of the moving object detected by the direction detection unit and the movement of the moving object detected by the motion detection unit;
A moving object detection apparatus for detecting a moving object from within an area predicted by the area prediction means. Imaging step;
A moving object detection step of detecting a moving object from the captured image captured in the imaging step;
A direction detecting step for detecting a direction of the moving object detected in the moving object detecting step;
A motion detection step of detecting the motion of the moving object from the position of the moving object detected from each captured image captured at different times in the imaging step;
A region prediction step of predicting a region where a moving object exists in a captured image based on the direction of the moving object detected in the direction detection step and the movement of the moving object detected in the motion detection step;
A moving object detection method, comprising: detecting a moving object from the area predicted in the area prediction step.

Images