Abnormal traveling detecting device

Abstract

To quickly and reliably determine an abnormal traveling vehicle which is present in a distant or nearby area.SOLUTION: An abnormal traveling detecting device 100 comprises: a learning unit 31 that obtains a vanishing point EX from an advance image P0 taken at a setting processing level in advance, and determines a first area A1 within a photographic range PA, which includes the vicinity of the vanishing point EX; a detection unit 36a that detects an image P1 with a low degree of compression in the first area A1 and an image P2 having a high degree of compression in a second area A2 including a neighboring NF side from a target image OI captured at a detection processing level; and a determination unit 36c that determines whether a vehicle is abnormally traveling based on the vehicle information VI.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an abnormal driving detection device that detects an abnormally driving vehicle that may cause an accident, for example. [Background technology]

[0002] A known obstacle detection device for detecting obstacles ahead of a vehicle calculates a vanishing point from road boundary information, generates an enlarged image by enlarging the area around the vanishing point, converts the previous enlarged image into a converted enlarged image according to the distance, and compares these to create a differential image, which is then binarized using a predetermined threshold value to detect the obstacle (Patent Document 1).

[0003] The device in Patent Document 1 does not detect dangerous driving, and even if it could detect obstacles, it would not necessarily be easy to apply it to dangerous driving detection. Furthermore, the method in Patent Document 1 cannot acquire clear images at night, etc., and there is a possibility that obstacles cannot be detected accurately.

[0004] It is also conceivable to determine dangerous driving by using AI image processing of images captured by cameras monitoring roads, but AI image processing places a heavy processing load on the system, and if real-time performance is required, it is necessary to either increase the hardware specifications or reduce the number of pixels to be processed by shrinking the image. If the image is reduced, it becomes difficult to detect vehicles at a distance. One method to solve the above problems would be to crop a portion of the image and make a judgment, but this requires setting the crop position, and in applications where the installation location changes frequently, it is time-consuming to change the settings according to changes in composition, and the knowledge required for setting up makes this method difficult to use. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-170568 Summary of the Invention [Problem to be solved by the invention]

[0006] The present invention has been made in view of the above-mentioned background art, and has as its object to quickly and reliably determine whether an abnormally moving vehicle is present in a distant area or a nearby area.

[0007] In order to achieve the above-mentioned object, the abnormal driving detection device of the present invention comprises a learning unit that, in a preliminary setting processing stage, finds a vanishing point from a captured preliminary image and determines a first area within the captured range that includes the vicinity of the vanishing point; a detection unit that, in a detection processing stage, detects vehicle information from an image with a low degree of compression in the first area of ​​the captured target image and an image with a high degree of compression in a second area of ​​the target image that includes the nearby side; and a judgment unit that judges whether or not the driving is abnormal based on the vehicle information.

[0008] In the above-mentioned abnormal driving detection device, the detection unit detects vehicle information from a low-compression image of a first region of the captured target image and a high-compression image of a second region of the target image that includes the nearby side during the detection processing stage, so that vehicle information can be obtained with high accuracy from the low-compression image for the first region that targets the distant area, and vehicle information can be obtained relatively quickly from the high-compression image for the second region that mainly targets the nearby area. This makes it possible to quickly and reliably monitor vehicles traveling on roads over a wide area from distant to nearby areas, and improves the reliability of abnormal driving determination. [Brief explanation of the drawings]

[0009] [Figure 1] 1A and 1B are block diagrams illustrating an abnormal driving detection device according to an embodiment. [Figure 2] FIG. 2 is a diagram showing an installation state of an abnormal driving detection device. [Figure 3] 1A is a conceptual diagram illustrating the operation of a vanishing point determination unit, and FIG. 1B is a conceptual diagram illustrating the operation of a driving lane determination unit. [Figure 4] 4 is a flowchart illustrating the operation of the abnormal driving detection device according to the embodiment. [Figure 5] 10 is a conceptual flowchart illustrating the configuration process steps. [Figure 6] 1 is a conceptual flow chart illustrating the detection process steps. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment of an abnormal driving detection device according to the present invention will be described with reference to the drawings.

[0011] Fig. 1(A) is a block diagram showing an example of the configuration of an abnormal driving detection device of an embodiment, and Fig. 1(B) is a functional block diagram explaining the abnormal driving detection device. Fig. 2 is a conceptual side view showing the installation state of the abnormal driving detection device 100 shown in Fig. 1. Fig. 3(A) is a conceptual diagram explaining the operation of the vanishing point determination unit, and Fig. 3(B) is a conceptual diagram explaining the operation of the driving lane determination unit.

[0012] Referring to Figure 1(A) etc., the abnormal driving detection device 100 includes a camera 10 that photographs the road RO (see Figure 2) on which the vehicle is traveling, and an information processing device 20 that makes a determination regarding the driving state of the vehicle from the photographed image PI (see Figure 1(B)) acquired by the camera 10.

[0013] The camera 10 has a shooting range PA (see FIG. 3) that is the target space including the target road RO, the area above it, and its surroundings, and continuously captures images of the target at a predetermined time interval (for example, about 30 fps). The captured images PI, including road images captured by the camera 10, are output to the information processing device 20. The frame rate of the continuous shooting by the camera 10 may be, for example, 5 fps or 10 fps, or may be about 100 fps or higher.

[0014] The information processing device 20 is a computer, and includes a main control device 21, a storage device 22, a user interface 23, and a communication device 24. The information processing device 20 detects abnormally driving vehicles that may cause an accident using the captured images PI captured by the camera 10. To this end, the information processing device 20 performs a setting process in advance and a detection process that detects vehicle information and determines whether the vehicle is driving abnormally based on this vehicle information, as will be described in detail later.

[0015] The main control device 21 operates based on a program stored in the storage device 22. The storage device 22 stores basic programs for operating the information processing device 20 and application software that operates on the basic programs. The application software includes software for operating the camera 10 to acquire captured images PI, software for executing a pre-setting processing step, and software for executing a detection processing step for detecting vehicle information from the captured images PI. The user interface 23 includes a display, touch panel, speaker, microphone, etc., and accepts commands from an operator and presents the operator with the processing results of the information processing device 20. The information processing device 20 can communicate with an external management server (not shown) via a communication network using the communication device 24.

[0016] 1(B), the information processing device 20 has, as parts for executing the setting processing stage, a learning unit 31, a detection image creation unit 34, and a user interface unit 35. The information processing device 20 also has, as parts for executing the detection processing stage, a detection processing unit 36, a detection image creation unit 34, and a user interface unit 35.

[0017] In the information processing device 20, the learning unit 31, in the setting processing stage, determines a vanishing point EX from a captured preliminary image and determines a relatively narrow first area within the shooting range PA that is part of the shooting range PA and includes the vicinity of the vanishing point EX (see FIG. 3, which will be described in detail later, for the shooting range PA, vanishing point EX, and first area A1). The learning unit 31 has an image processing unit 31a, a vanishing point determination unit 31b, an area determination unit 31c, and a driving lane determination unit 31d. In the detection processing stage, the detection processing unit 36 ​​detects vehicle information from the captured target image and determines whether or not the vehicle is driving abnormally based on the vehicle information. The detection processing unit 36 ​​has a detection unit 36a and a determination unit 36c.

[0018] Referring to FIG. 2, the abnormal driving detection device 100 is mounted on a sign vehicle V1 located adjacent to a maintenance work site MW where construction work and fallen objects are being carried out on a road RO, such as a highway. The abnormal driving detection device 100 includes a camera 10 fixed to a sign 3 fixed to a loading platform 2a, an information processing device main body 20a fixed to the loading platform 2a, and a user interface 23 installed in the driver's seat 2b. The maintenance work site MW is located in front of the sign vehicle V1. A human-shaped warning sign 4 is located behind the sign vehicle V1, upstream in the vehicle's travel direction, and supports a rotating warning light 4a. Upstream of the human-shaped warning sign 4 in the vehicle's travel direction, a vehicle V2 is traveling on the road RO and approaching the maintenance work site MW. The sign 3 urges the vehicle V2 to slow down and change lanes. Note that the warning sign 4 does not have to be a human-shaped sign; it may be the same as the sign on the sign vehicle V1 or may be replaced with another sign.

[0019] 2 and 3(A), a target area including the road RO behind the sign vehicle V1 and the upstream direction of the vehicle is photographed by the camera 10. The photographing range PA of the camera 10 includes the outside of the road RO, but it is not essential that the vanishing point EX, which traces the trajectory of the vehicle V2 and the like in the reverse direction, be included in the angle of view. The photographed image PI, which includes a road image output from the camera 10, is processed by the information processing device main body 20a, and the processing result is output to the user interface 23.

[0020] Returning to FIG. 1B etc., in the information processing device 20, the image processing unit 31a of the learning unit 31 operates the camera 10 during the setting process to acquire an image of the road RO ahead and the area above it as a preliminary image P0, and extracts one or more images of the vehicle V2 from this preliminary image P0. Known pattern recognition techniques can be used to extract the vehicle V2. By incorporating pattern recognition software trained by machine learning, such as deep learning, into the image processing unit 31a as a trained model for vehicle extraction, the image area of ​​the vehicle V2 can be determined from the preliminary image P0 and the front area (pixel position) of the vehicle V2 or its center coordinates can be calculated. Any type of pattern recognition software can be used as long as it can recognize the vehicle V2. However, distinguishing attributes such as the vehicle type and color makes it easier to check the identity of the extracted vehicle images. A specific example of pattern recognition software is one that uses, for example, histograms of oriented gradients (HOG) to extract features and, for example, a support vector machine as a learning model for identifying the features. The method for extracting vehicle images is not limited to the above-described machine learning method, but may also be a method for extracting artificially set local features using a statistical learning method, such as optical flow or template matching.

[0021] The image processing unit 31a receives the captured images PI from the camera 10 as successive frames (e.g., 10 fps, 30 fsp) and sequentially extracts vehicle images from each captured image PI. The image processing unit 31a determines the pixel position or screen coordinates of the vehicle image in each of the successive captured images PI over time. By tracking a specific vehicle image, the image processing unit 31a can determine the movement vector of the vehicle V2 as the positional deviation of the vehicle image between the captured images PI, and obtain the movement trajectory of the vehicle V2. At night, the images from the camera 10 mainly include headlights. Even in such a case, the image processing unit 31a can identify and track the vehicle V2 using pattern recognition technology.

[0022] The image processing unit 31a can set a rectangular speed measurement area along the road RO in the captured image PI, monitor a vehicle V2 passing through this speed measurement area, and determine the speed of the vehicle V2 from the time it passes through. The size of the speed measurement area, i.e., the actual distance, is known. The speed measurement area can be set by setting four reference points on the road RO in advance. Information on the world coordinates (latitude and longitude) of the reference points is acquired, and the reference points can be set by identifying an object in the captured image PI, placing an actual indicator, or marking the captured image PI. Note that instead of the speed measurement area, a speed measurement start line and a speed measurement end line set to cross the road RO can also be used. Two or more speed measurement areas can be set in the captured image PI.

[0023] The learning unit 31 performs pre-learning as a setting processing stage under the supervision of an operator. Based on instructions from the operator, the learning unit 31 adjusts the attitude of the camera 10 and sets a shooting range PA that includes the road RO. The learning unit 31 accumulates information while learning from captured images PI obtained by capturing images of vehicles V2 actually passing through the road RO. The number of vehicles V2 to be captured is, for example, between 10 and 100, but the number of vehicles captured in advance can be increased or decreased depending on the situation. At this time, the speed of the vehicles V2 can be measured and vehicles V2 that are exceeding the speed limit can be excluded from the learning target.

[0024] The vanishing point determination unit 31b of the learning unit 31 determines a vanishing point EX based on the movement trajectories VT obtained by the image processing unit 31a for multiple vehicles V2. The vanishing point determination unit 31b has the function of extrapolating or extending the upper ends of the movement trajectories VT, and determines the vanishing point EX as the position where the extension lines of the multiple movement trajectories VT intersect or the position where the extension lines are closest to each other. The extrapolation of the upper ends of the movement trajectories VT may be either a straight line or a curve, or may be extended using an approximation curve. The vanishing point EX does not necessarily exist within the captured image PI, but may also exist outside the captured image PI, and is specified by screen coordinates.

[0025] The operation of the vanishing point determination unit 31b will be described with reference to FIG. 3(A). A road RO with two lanes is captured in the captured image PI, which is the preliminary image P0. The image processing unit 31a detects multiple movement trajectories VT acquired for multiple vehicles V2. These movement trajectories VT are virtual, not images of real objects, but can be superimposed on the captured image PI on the display of the user interface unit 35. Multiple extension lines EL, shown as dotted lines, extend above the movement trajectories VT, intersecting or approaching each other. The vanishing point determination unit 31b determines the vanishing point EX as the position where the multiple extension lines EL intersect or the position where the extension lines EL are closest to each other. The image used to determine the vanishing point EX may be, but is not limited to, the uncompressed captured image PI as it was captured. The determination of the vanishing point EX is intended to determine the first region A1, and does not need to be highly accurate. Using a compressed image makes the pre-processing stage efficient and rapid. If a compressed image is used to determine the vanishing point EX, the image can be compressed as part of image processing, but a compressed image may also be prepared separately. In addition, in this embodiment, the vanishing point EX is determined as the point where the extension line EL intersects with the movement trajectory VT of the vehicle V2, so it is possible to deal with cases where the detection accuracy is low in the distance.

[0026] 1(B) and 3(A), the region determination unit 31c of the learning unit 31 determines a first region A1 including the vicinity NB of the vanishing point EX in the captured image PI, which is the preliminary image P0, i.e., within the capturing range PA. The image of the first region A1 is used to measure the distant portion of the road RO during the detection processing stage or the setting processing stage. The first region A1 fits within a preset image or pixel size (M × N pixels or M × M pixels) and is, for example, approximately half or less the image size of the entire capturing range PA. The image size of the first region A1 is, for example, 300 × 300 pixels, but can be changed as appropriate depending on the application and installation environment of the abnormal driving detection device 100. When the vanishing point EX is within the capturing range PA, the first region A1 is set to be positioned as far below the capturing range PA as possible, provided that it includes a predetermined number of pixels surrounding the vanishing point EX. This allows a large area to be secured below the vanishing point EX in the first region A1. When the vanishing point EX is outside the shooting range PA, i.e., above it, the first area A1 is set to be located at the top of the shooting range PA. This causes the first area A1 to include at its top an image corresponding to the point on the road RO that is closest to the vanishing point EX.

[0027] The region determination unit 31c determines a second region A2 including the first region A1 within the captured image PI, i.e., the capturing range PA. The image of the second region A2 is used to measure the near portion of the road RO in the detection processing stage or the setting processing stage. The second region A2 includes the neighboring NF side of the first region A1 within the capturing range PA. The second region A2 is usually the entire capturing range PA, but can be the region excluding the outside of the road RO. The second region A2 may also be the capturing range PA excluding the first region A1.

[0028] The lane determination unit 31d accumulates the movement trajectories VT of the number of vehicles V2 captured in advance through advance learning as information, and determines a frame-shaped lane TZ0 and a no-travel area FA (FIG. 3B) that encompass these. The operation of the lane determination unit 31d will be described in detail later.

[0029] The detection image creation unit 34 creates a low-compression image P1 from an image IM1 in the first region A1 of the captured image PI acquired by the camera 10, and creates a high-compression image P2 by appropriately compressing an image IM2 in the second region A2 of the captured image PI. The low-compression image P1 created by the detection image creation unit 34 is specifically an uncompressed image obtained by simply cropping out a portion of the original captured image PI, and fits within a range of, for example, 300 x 300 pixels. The high-compression image P2 created by the detection image creation unit 34 is a compressed image obtained by compressing image IM2 in the captured image PI (in this specific example, the entire captured image PI) by, for example, 25%, and fits within a range of, for example, 300 x 300 pixels. The low-compression image P1 is accompanied by transformation information that allows two-dimensional coordinates on the original captured image PI to be reproduced. Similarly, the highly-compressed image P2 is accompanied by transformation information that allows two-dimensional coordinates on the original captured image PI to be reproduced. This allows the movement vector and movement trajectory VT of vehicle V2 obtained from image P1 with a low degree of compression and the movement vector and movement trajectory VT of vehicle V2 obtained from image P2 with a high degree of compression to be managed using common coordinates, thereby making this information consistent.

[0030] The driving lane determination unit 31d receives a detection image PD (i.e., images P1 and P2) created by processing the preliminary image P0 taken during the setting processing stage by the detection image creation unit 34, and determines the driving locus zone TZ0 based on this detection image PD.

[0031] The operation of the travel lane determination unit 31d will be described with reference to Figures 1(B) and 3(B). In the photographed image PI, which is the preliminary image P0, a travel locus zone TZ0 is set along the road RO. This travel locus zone TZ0 is virtual, but can be displayed superimposed on the photographed image PI on the display of the user interface unit 35.

[0032] The travel locus zone TZ0 can be expanded laterally within the photographing range PA. In this case, the expanded buffer area BA is collectively referred to as the travel locus zone TZ1. The buffer area BA is set, for example, by an operator, but may also be determined by the travel lane determination unit 31d based on predetermined criteria. The travel locus zones TZ0 and TZ1 serve as criteria for determining whether the vehicle V2 is traveling abnormally. As will be described later, if the center or both front ends of the vehicle V2 extend beyond the travel locus zones TZ0 and TZ1 and go outside the travel locus zones TZ0 and TZ1, the detection processing unit 36 ​​determines that the vehicle V2 is traveling abnormally.

[0033] The lane determination unit 31d can set a no-travel area FA based on instructions from an operator, etc. The no-travel area FA can be set arbitrarily from the perspective of ensuring safety, and in the example of FIG. 2, it is set in the area from the marked vehicle V1 to the warning sign 4.

[0034] The learning unit 31 automatically determines or accepts from an operator a criterion for determining whether the target vehicle V2 is driving abnormally by how much it deviates from the driving locus zones TZ0, TZ1. The learning unit 31 automatically determines or accepts from an operator a criterion for determining whether the target vehicle V2 is driving abnormally by how much it meanders. Here, whether or not the meandering is excessive is not directly related to whether or not the target vehicle V2 deviates from the driving locus zones TZ0, TZ1; for example, the target vehicle V2 may be determined to be driving abnormally even if it does not deviate from the driving locus zones TZ0, TZ1. The learning unit 31 automatically determines or accepts from an operator a maximum speed limit for determining whether or not the target vehicle V2's speed is abnormal.

[0035] Referring to Figure 1(B), in the detection processing stage, the detection unit 36a of the detection processing unit 36 ​​operates the camera 10 to capture a target image OI with the same angle of view as the preliminary image P0, and operates the detection image creation unit 34 to create a low-compression image P1 from an image IM1 in a first region A1 (see Figure 3(A)) of the captured image PI, which is the target image OI, and creates a highly compressed image P2 by appropriately compressing an image IM2 in a second region A2 (see Figure 3(A)) of the captured image PI.

[0036] The detection unit 36a includes an image processing unit 36b with functions similar to those of the image processing unit 31a of the learning unit 31. During the detection processing stage, the image processing unit 36b performs image processing on the low-compression image P1 and the high-compression image P2 created by the detection image creation unit 34 to detect vehicle information VI. Specifically, the image processing unit 36b extracts one or more images of the vehicle V2 based on the images P1 and P2 obtained from the detection images PD. The image processing unit 36b determines the pixel positions or screen coordinates of the vehicle images in the sequential images P1 and P2 over time, tracks a specific vehicle image, and determines the movement vector of the vehicle V2, thereby obtaining the movement trajectory VT of the vehicle V2. Here, the images P1 and P2 have different compression levels but are managed using common coordinates, and the movement vector and movement trajectory VT of the vehicle V2 are managed in a single coordinate system as the detection images PD. In addition, the image processing unit 36b can set a rectangular speed measurement area along the road RO within the images P1 and P2, monitor the vehicle V2 passing through this speed measurement area, and determine the speed of the vehicle V2 from the time it takes to pass through.

[0037] The determination unit 36c determines whether the vehicle V2 is traveling abnormally based on the vehicle information VI obtained by the detection unit 36a, specifically the movement trajectory VT and speed of the vehicle V2. The determination unit 36c determines that the vehicle V2 is traveling abnormally if: (1) the vehicle V2 of interest travels beyond the travel trajectory zones TZ0 and TZ1 by a distance greater than a predetermined distance; (2) the vehicle V2 of interest travels within the travel trajectory zones TZ0 and TZ1 while meandering significantly; or (3) the vehicle V2 of interest exceeds a predetermined upper speed limit. Note that the "distance greater than a predetermined distance" is based on pixel units, and the horizontal pixel width differs between the far side and the near side. Furthermore, the "distance greater than a predetermined distance" can be set by an operator in the pre-processing stage (during pre-learning), but may also be automatically determined from the horizontal width of the travel trajectory zones TZ0 and TZ1 when determining the travel trajectory zones TZ0 and TZ1.

[0038] FIG. 4 is a flowchart illustrating the operation of the abnormal driving detection device 100 shown in FIGS. 1(A) and 1(B). The abnormal driving detection device 100 first checks whether it is in the setting processing stage (step S01). If it is in the setting processing stage (YES in step S01), the abnormal driving detection device 100 performs setting processing (step S02). If it is not in the setting processing stage (NO in step S01), that is, if it is in detection processing, the abnormal driving detection device 100 performs detection processing (step S03). If it has performed detection processing (step S03), the abnormal driving detection device 100 outputs the detection result to the user interface unit 35 or to the outside (step S04). Thereafter, the abnormal driving detection device 100 checks whether it should end the processing (step S05). If it is not to end the processing (NO in step S05), it returns to step S01 to check whether it is in the setting processing stage (step S01), and then repeats the same processing.

[0039] FIG. 5 is a flowchart illustrating the setting process. The learning unit 31 checks whether the calculation of the vanishing point EX is complete (step S11). If the calculation of the vanishing point EX is not complete (NO in step S11), the learning unit 31 starts inputting a camera image capturing a captured image PI, i.e., a preliminary image P0, from the camera 10 (step S12). The captured image PI thus acquired corresponds to the original data not processed by the detection image creation unit 34. The image processing unit 31a of the learning unit 31 performs image processing on the captured image PI to acquire vehicle movement trajectories VT (step S13). The learning unit 31 continues inputting camera images until movement trajectories VT are obtained for a predetermined number of vehicles or until an instruction is received from the operator. Thereafter, the vanishing point determination unit 31b of the learning unit 31 calculates the vanishing point EX based on the movement trajectories VT acquired so far (step S14).

[0040] When the calculation of the vanishing point EX is completed (YES in step S11), the learning unit 31 checks whether the area determination for image processing is completed (step S15). When the area determination is not completed (NO in step S15), the area determination unit 31c of the learning unit 31 determines a first area A1 far from the shooting range PA based on the vanishing point EX (step S16), and determines a second area A2 including the nearby NF side (step S17). The second area A2 is the entire shooting range PA, but is not limited to this.

[0041] When the image processing region determination is completed (YES in step S15), the learning unit 31 causes the detection image creation unit 34 to process the captured image PI and create a detection image PD (i.e., images P1 and P2) (step S18). Thereafter, the image processing unit 31a of the learning unit 31 performs image processing on image P1 of the detection images PD to obtain a vehicle movement trajectory VT (step S19), and performs image processing on image P2 of the detection images PD to obtain a vehicle movement trajectory VT (step S20). Note that the image processing unit 31a reuses the captured image PI, i.e., the preliminary image P0, obtained in step S12. At this time, for image P1 with a low compression level, the image processing unit 31a can extract a vehicle using a model for distant detection and determine a vehicle movement trajectory VT. For image P2 with a high compression level, the image processing unit 31a can extract a vehicle using a model for non-distant detection and determine a vehicle movement trajectory VT. Thereafter, the learning unit 31 acquires the processing results based on the vehicle movement trajectory VT, etc. as the learning result (step S21). Specifically, the learning unit 31 integrates the movement trajectory VT obtained in step S19 and the movement trajectory VT obtained in step S20 as vehicle information to obtain a continuous movement trajectory VT at coordinates on the detection image PD. Such integrated movement trajectories VT are obtained for a large number of vehicles and are stored as information in the learning unit 31. The movement trajectory VT is a movement trajectory zone that encompasses normal movement trajectories. The movement lane determination unit 31d of the learning unit 31 determines a frame-shaped movement trajectory zone TZ0 or a no-travel area FA that encompasses a large number of integrated movement trajectories VT. Furthermore, the learning unit 31 automatically determines the conditions for abnormal driving or accepts designation from the operator. As mentioned above, abnormal driving is determined to occur when (1) the vehicle in question is driving at a distance greater than a predetermined distance from the driving locus zones TZ0 and TZ1, (2) the vehicle in question is driving while meandering significantly even within the driving locus zones TZ0 and TZ1, or (3) the vehicle in question exceeds a predetermined upper speed limit.

[0042] The learning unit 31 confirms that the learning has been completed (step S22), and if the learning has not been completed, returns to step S11 and repeats the processing of steps S11 to S22, and if the learning has been completed, ends the setting processing.

[0043] FIG. 6 is a flowchart illustrating the detection processing stage. The detection unit 36a of the detection processing unit 36 ​​starts inputting a camera image capturing a captured image PI, i.e., a target image OI, from the camera 10 (step S31). The detection unit 36a causes the detection image creation unit 34 to process the captured image PI and create a detection image PD (i.e., images P1 and P2) (step S32). Thereafter, the detection unit 36a of the detection processing unit 36 ​​performs image processing on image P1 of the detection images PD to obtain first vehicle information (step S33), and performs image processing on image P2 of the detection images PD to obtain second vehicle information (step S34). At this time, the image processing unit 36b of the detection unit 36a can extract vehicles from image P1, which has a low degree of compression, using a model for distant detection and determine a vehicle movement trajectory VT. Meanwhile, the image processing unit 36b can extract vehicles from image P2, which has a high degree of compression, using a model for non-distant detection and determine a vehicle movement trajectory VT. Thereafter, the detection unit 36a integrates the first vehicle information obtained in step S33 and the second vehicle information obtained in step S34 (step S35). That is, the detection results for the first area A1 and the detection results for the second area A2 are managed in a common coordinate system. The detection unit 36a determines the coordinate position of the vehicle while checking the identity of the extracted vehicle images based on the information integrated in step S35 (step S36). If multiple vehicles are present in the captured image PI, an ID is assigned to each vehicle and the coordinate position is managed. At this time, a movement vector is obtained for each vehicle. If the speed of the vehicle can be verified using a speed measurement area set along the road RO in the images P1 and P2, the detection unit 36a determines the speed of the vehicle (step S37). The detection unit 36a outputs vehicle information VI (ID, coordinate position, speed vector, etc.) including the coordinate position and speed of the vehicle to the determination unit 36c of the detection processing unit 36. The determination unit 36c determines whether the vehicle is traveling abnormally based on the vehicle information VI received from the detection unit 36a, i.e., the coordinate position of the vehicle obtained in step S36 and the vehicle speed obtained in step S37 (step S38). If a vehicle determined to be traveling abnormally is present, the detection processing unit 36 ​​notifies the operator of the presence of the abnormally traveling vehicle together with the detection information via the user interface unit 35.The notification of an abnormally driving vehicle includes vehicle images and audio. The detection information can include vehicle information VI, abnormal driving determination results, detection time, etc. The detection processing unit 36 ​​can also output the detection information and corresponding commands to an external device. This allows the display, which is the sign 3 of the marked vehicle V1 shown in Figure 2, to display a warning such as "Abnormally Driving Vehicle Present" to the abnormally driving vehicle and surrounding vehicles.

[0044] The abnormal driving detection device 100 of the embodiment described above includes a learning unit 31 that, in a preliminary setting processing stage, finds a vanishing point EX from the captured preliminary image P0 and determines a first area A1 within the shooting range PA that includes the vicinity of the vanishing point EX; a detection unit 36a that, in the detection processing stage, detects vehicle information VI from an image P1 with a low degree of compression in the first area A1 of the captured target image OI and an image P2 with a high degree of compression in a second area A2 of the target image OI that includes the neighboring NF side; and a judgment unit 36c that judges whether or not the driving is abnormal based on the vehicle information VI.

[0045] In the above-described abnormal driving detection device 100, the detection unit 36a detects vehicle information VI during the detection processing stage from a low-compression image P1 in a first region A1 of the captured target image OI and a highly-compressed image P2 in a second region A2 of the target image OI that includes the nearby NF. Therefore, for the first region A1 that targets the distant area, vehicle information VI can be obtained with high accuracy from the low-compression image P1, and for the second region A2 that targets mainly the nearby NF, vehicle information VI can be obtained relatively quickly from the highly-compressed image P2. This makes it possible to quickly and reliably monitor vehicles traveling on roads RO over a wide area, from distant to nearby regions, and improve the reliability of abnormal driving determination.

[0046] 〔others〕 The present invention is not limited to the above-described embodiment, and can be embodied in various forms without departing from the spirit and scope of the present invention.

[0047] For example, the detection image creation unit 34 creates an uncompressed image for image IM1 in the first area A1 of the captured image PI, but may create image P1 with a relatively lower compression level than image IM2 in the second area A2.

[0048] The detection image creation unit 34 determines the first area A1 and the second area A2 from the captured image PI to obtain the low-compression image P1 and the high-compression image P2, but a third area may be provided adjacent to and below the first area A1. In this case, the image with the lowest compression level is obtained from the first area A1, the image with an intermediate compression level is obtained from the third area, and the image with the highest compression level is obtained from the second area A2.

[0049] The image processing unit 31a of the learning unit 31 is not limited to one that uses trained pattern recognition software, but may be one that is capable of learning to improve compatibility on-site.

[0050] The abnormal driving detection device 100 and the camera 10 do not have to be mounted on the sign vehicle V1. For example, one or more cameras 10 can be installed near or at a remote location from the sign vehicle V1 to detect abnormal driving vehicles remotely or simultaneously in parallel. Note that if the camera 10 is installed at a location remote from the information processing device 20, a communication circuit will be required to remotely operate the camera 10 and its associated drive circuit, and a support stand will also be required to set the camera 10 at a relatively high viewpoint position.

[0051] In the above explanation, the speed measurement area is used to verify the vehicle speed. This corresponds to determining the vehicle speed by inputting the latitude and longitude of several points on the captured image PI screen and calculating the distance. The present invention is not limited to this measurement method. For example, an operator can operate multiple points on the captured image PI screen captured by the camera 10 to provide relative distance (distance) information, and the vehicle speed can be determined from this distance information. The vehicle speed can also be determined by inputting the distance from the camera 10 to a required point on the captured image PI screen. The vehicle speed can also be determined by calculating the distance from something that specifies the size of the vehicle (for example, the size of the license plate). In addition, a determination similar to the vehicle speed can be made using the moving speed on the screen obtained by measuring the time it takes to pass through a section set in the captured image PI. [Explanation of symbols]

[0052] 3...sign, 10...camera, 20...information processing device, 20a...information processing device main body, 21...main control device, 22...storage device, 23...user interface, 24...communication device, 31...learning unit, 31a...image processing unit, 31b...vanishing point determination unit, 31c...area determination unit, 31d...driving lane determination unit, 34...output image creation unit, 35...user interface unit, 36...detection processing unit, 36a...detection unit, 36b...image processing unit, 36c...judgment unit, 100...abnormal driving detection device, BA...buffer area, EL...extension line, EX...vanishing point, FA...no driving area, MW...maintenance work site, NB...vicinity, OI...target image, P0...preliminary image, P1...low compression image, P2...high compression image, PA...shooting range, PD...detection image, PI...shooting image, RO...road, TZ0, TZ1...Travel trajectory, V1...Vehicle sign, V2...Vehicle, VI...Vehicle information, VT...Trajectory

Claims

1. a learning unit that, in a setting processing stage performed in advance, obtains a vanishing point from a captured preliminary image and determines a first area including a vicinity of the vanishing point within the captured range; a detection unit that detects vehicle information from an image having a low compression degree in the first region of the captured target image and an image having a high compression degree in a second region including a nearby side of the captured target image in a detection processing stage; a determination unit that determines whether or not the vehicle is traveling abnormally based on the vehicle information; An abnormal driving detection device comprising:

2. The abnormal movement detection device according to claim 1 , wherein the image with a low degree of compression in the first region is an image that is not compressed in the first region.

3. The abnormal driving detection device according to claim 1 , wherein the learning unit extracts a vehicle from the preliminary image and determines the vanishing point from a movement trajectory of the vehicle.

4. The abnormal driving detection device according to claim 3 , wherein the learning unit determines the vanishing point by extracting a vehicle from an image obtained by compressing the preliminary image.

5. The abnormal driving detection device according to claim 1 , wherein the second region includes the entire captured target image.

6. 2. The abnormal driving detection device according to claim 1, wherein the detection unit detects vehicle information from the image with a low compression degree of the first region and the image with a high compression degree of the second region using a trained model for vehicle extraction.

7. The abnormal driving detection device according to claim 6 , wherein the detection unit distinguishes between vehicles by integrating the vehicle information obtained in the first area and the vehicle information obtained in the second area.

8. The abnormal driving detection device according to claim 3 , wherein the learning unit determines a normal driving locus zone from the movement locus.

9. 9. The abnormal driving detection device according to claim 8, wherein the determination unit determines that the vehicle is driving abnormally when the coordinates of the vehicle included in the vehicle information deviate from the normal driving locus by a predetermined amount or more.

10. The abnormal driving detection device according to claim 8 , wherein the determination unit determines that the vehicle is driving abnormally when the vehicle speed included in the vehicle information is equal to or greater than a predetermined speed.

Images