A refined traffic incident detection method at intersections based on edge computing sensing and multi-source information fusion
By employing a hierarchical processing mechanism that enables rapid perception from edge computing devices and in-depth analysis from the cloud, the problem of real-time response and accurate identification of traffic incidents at intersections has been solved. This achieves a balance between real-time performance and accuracy in traffic incident handling, while reducing network transmission pressure and the frequency of large model service calls.
Patent Information
- Application Number
- CN202611351300.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-02
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-09-02
AI Technical Summary
Existing technologies struggle to achieve a balance between real-time response, accurate identification, and resource efficiency for traffic incidents at intersections under limited bandwidth and computing power, especially in complex traffic scenarios where fine-grained event classification and cause determination are difficult.
A hierarchical processing mechanism is adopted, which combines rapid perception by edge computing devices with in-depth analysis by the cloud. The video stream from the traffic camera is acquired by the edge computing device, and image frame processing and target detection are performed. The vehicle movement speed is calculated by combining the perspective transformation matrix, a JSON array is constructed and sent to downstream applications or databases, and the cloud performs fine-grained judgment.
It achieves a balance between real-time performance and accuracy in handling traffic events, reduces network transmission pressure and the frequency of large model service calls, and provides reliable technical support for intelligent transportation systems.
Smart Images

Figure CN122867419B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation technology, specifically to a refined traffic incident detection method for intersections based on edge computing perception and multi-source information fusion. Background Technology
[0002] With the accelerating pace of urbanization and the increasing complexity of road traffic systems, real-time and accurate detection of traffic incidents at intersections has become a key requirement in the field of intelligent transportation. Traditional traffic incident detection mainly relies on fixed sensors, geomagnetic coils, or centralized video analysis systems. These methods often suffer from high deployment and maintenance costs, long response delays, and coarse-grained recognition, making it difficult to meet the real-time perception and management needs of modern intersections for refined events such as abnormal parking, overflow congestion, and traffic accidents.
[0003] In recent years, computer vision-based traffic incident detection technology has become increasingly widespread, with edge computing and cloud-based collaborative processing architectures attracting significant attention. Edge computing, by performing preliminary calculations on devices close to the data source, can significantly reduce data transmission latency and achieve real-time basic incident detection. However, limited by the computing power and storage resources of edge devices, existing edge detection models often have limited accuracy, especially in complex traffic scenarios where fine-grained event classification and cause determination are difficult.
[0004] On the other hand, artificial intelligence technologies, represented by large-scale visual models, possess powerful feature extraction and semantic understanding capabilities, demonstrating excellent performance in refined image analysis. While directly transmitting all video streams to the cloud and relying on large models for full analysis could improve recognition accuracy, it would also place enormous pressure on network transmission and incur high service call costs, while making it difficult to guarantee the real-time nature of event detection.
[0005] Therefore, existing technologies lack a solution for intersection traffic event detection that can balance real-time response, accurate identification, and resource efficiency. How to effectively combine the real-time advantages of edge computing with the refined capabilities of large-scale model inference to achieve hierarchical and accurate detection of intersection traffic events under limited bandwidth and computing power has become a pressing technical problem in this field. Summary of the Invention
[0006] To overcome the shortcomings of the above technologies, this invention provides a hierarchical processing mechanism that combines rapid edge perception with in-depth cloud analysis to achieve a balance between traffic incident detection efficiency and accuracy, thus providing reliable technical support for real-time decision-making and intervention in intelligent transportation systems. This method is based on edge computing perception and multi-source information fusion for refined traffic incident detection at intersections.
[0007] The technical solution adopted by this invention to overcome its technical problems is: A refined traffic event detection method for intersections based on edge computing perception and multi-source information fusion includes: S1. Obtain the video stream information from the traffic enforcement camera at the intersection, and process the video into frames to obtain... Image set of intersection road images , ,in, For the first Frame intersection road image, ; S2. In the Frame intersection road image The parameters for determining the overflow area, the straight overflow area, the left turn overflow area, the overflow area, the overflow area, the straight overflow area, and the left turn overflow area are manually calibrated. S3. In the Frame intersection road image The four corner points of the perspective transformation region of the quadrilateral are manually marked. S4. Obtain the perspective transformation matrix using the four corner points of the perspective transformation region. ; S5. Regarding the first Frame intersection road image The system performs target detection and tracking on motor vehicles to obtain the target's coordinates, the width of the detection box, and the height of the detection box. S6. Using the perspective transformation matrix The coordinates of the target, the width of the detection box, and the height of the detection box are calculated to determine the target's position in the first place. Frame motion rate; S7. Utilize the target in the first Construct a JSON array containing the frame's motion rate, impending overflow region, straight overflow region, left turn overflow region, parameters for determining the impending overflow region, parameters for determining the straight overflow region, and parameters for determining the left turn overflow region. ; S8. Convert the JSON array Send it to downstream applications or store it in a database.
[0008] Furthermore, step S1 includes the following steps: S1-1. The edge computing device E-server obtains data from the intersection's electronic police camera via the RTSP protocol. Frame video information; S1-2. Using the Ascend Image Processing Unit (DVPP) in C++, the video information stream is decoded frame by frame into BGR format images. The cv::equalizeHist function from the OpenCV library is used to evaluate the first frame. The BGR format image of the frame is enhanced for contrast to obtain the first frame. Frame intersection road image .
[0009] Furthermore, step S2 includes the following steps: S2-1. In the first Frame intersection road image The lower left corner of the quadrilateral's overflow area is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is Construct a distal line parallel to the edge line of the zebra crossing at the front of the opposite exit lane, 20-40 meters along its length. Use the intersection of the left lane line of the opposite exit lane and the distal line as the upper left corner of the quadrilateral's overflow area. The x-coordinate of this point is... The ordinate of this point is The intersection of the right lane line and the far lane line of the opposite exit lane is taken as the upper right corner of the quadrilateral structure's overflow area, and the x-coordinate of this point is... The ordinate of this point is The lower right corner of the quadrilateral's overflow area is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S2-2. Construct a first proximal line parallel to the edge line of the zebra crossing at the front end of the opposite exit lane, 20-40 meters behind the edge line along its length. Frame intersection road image The lower left corner of the straight overflow area of the quadrilateral structure is defined by the intersection of the extension line of the left lane of the opposite exit lane and the first near-end line. The x-coordinate of this point is: The ordinate of this point is The upper left corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The upper right corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The intersection of the extension line of the right lane of the opposite exit lane and the first near-end line is taken as the lower right corner of the straight overflow area of the quadrilateral structure, and the x-coordinate of this point is . The ordinate of this point is ; S2-3. At the front end of the zebra crossing of the opposite exit lane, construct a second near-end line parallel to its edge line, 5 meters behind it along its length. At 10-20 meters behind the second near-end line, construct a third near-end line parallel to it. At 5 meters to the left of the left end of the left lane line of the opposite entrance lane, construct a left-end line parallel to it. Frame intersection road image The intersection of the third proximal line and the left end line is taken as the lower left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second proximal line and the left end line is taken as the upper left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second near-end line and the right lane line of the opposite entrance lane is taken as the upper right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the third near-end line and the right lane line of the opposite entrance lane is taken as the lower right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is ; S2-4. Manually calibrating the parameters for determining the overflow area. , ,in, To determine the threshold for vehicles to linger in the area that is about to overflow, To determine the speed threshold for vehicles to slow down in the area that is about to overflow, To determine the equivalent number of slow-moving standard vehicles in the area about to overflow; S2-5. Parameters for manually calibrating the straight-line overflow area , ,in, To determine the threshold for vehicles to linger in the straight-through overflow area, To determine the speed threshold for slow-moving vehicles in the straight-through overflow area, To determine the equivalent number of slow-moving standard vehicles in the straight-through overflow area; S2-6. Parameters for manually calibrating the left-turn overflow area , ,in, To determine the threshold for vehicles to linger in the left-turn overflow area, To determine the speed threshold for slow-moving vehicles in the left-turn overflow area, To determine the equivalent number of slow-moving standard vehicles in the left-turn overflow area.
[0010] Preferred, The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 2.
[0011] Furthermore, step S3 includes the following steps: S3-1. In the... Frame intersection road image The lower left corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite entrance lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S3-2. Use a GPS locator at the intersection to locate the front edge of the zebra crossing along its length at the front end of the opposite exit lane. Mark a real line parallel to the edge line at that point. The value ranges from 20 to 50 meters, in the... Frame intersection road image The intersection of the left lane line of the opposite entrance lane and the real line is taken as the upper left corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is ; S3-3. In the... Frame intersection road image The upper right corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S3-4. No. Frame intersection road image The intersection of the right lane line of the opposite exit lane and the real line is taken as the lower right corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is .
[0012] Furthermore, in step S4, the function cv::getPerspectiveTransform from the OpenCV library is used to obtain the coordinates of the four corner points of the perspective transformation region of the quadrilateral structure. , , , ,length Width of the exit lane for the target Convert to a 3x3 perspective transformation matrix .
[0013] Furthermore, step S5 includes the following steps: S5-1. Using the YOLOv11m model to analyze the... Frame intersection road image Target detection is performed on motor vehicles in the middle, and the results are obtained. The nth rectangular target detection box, the nth The x-coordinate of the center point of the detection box of each target is The center point's ordinate is Width is Height is , ; S5-2. Use the Deepsort tracking algorithm to track each detected target.
[0014] Furthermore, step S6 includes the following steps: S6-1. Through formula Calculation yields the first Information of the detection target in a real-world scenario ,in, For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. These are the perspective transformation coefficients; S6-2. Through formula Calculation yields the first The detection target is in the first Motion rate in a frame intersection road image In the formula, For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. This refers to the frame rate of the video from the intersection's electronic police camera, which is parsed by the edge computing device E-server via the RTSP protocol.
[0015] Furthermore, step S7 includes the following steps: S7-1. Using the vector cross product method to utilize the first Coordinates of the center point of the detection bounding box of each target Judge the first Whether the detection box of the first target is in the overflow area, the straight overflow area, or the left turn overflow area, if the first target is in the overflow area. If a detected target is in the overflow region, record the time the target spends in the overflow region. If the first If a detected target is in the straight-line overflow area, record the dwell time of the detected target in the straight-line overflow area. If the first If a detected target is in the left-turn overflow area, record the dwell time of the detected target in the left-turn overflow area. ; S7-2. If and Then determine the first [item] in the region that is about to overflow. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the impending overflow area among the detected targets is denoted as . ; S7-3. If and Then determine the first in the straight overflow region. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the straight-through overflow area among the detected targets is denoted as follows: ; S7-4. If and Then determine the first one in the left turn overflow area. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the left-turn overflow area among the detected targets is denoted as follows: ; S7-5. Creating Fields ,if and and This indicates normal. ,if and and This indicates that an overflow is imminent. ,if and This indicates straight overflow. ,if and This indicates a left turn overflow. ,if and and This indicates abnormal parking. ; S7-6. If or or , will the field , , , Store in JSON array middle.
[0016] Furthermore, the JSON array in step S8 Send the data to downstream applications or store it in a database via the Kafka service.
[0017] The beneficial effects of this invention are as follows: Edge computing devices are used to acquire video footage from electronic traffic enforcement cameras, and machine vision algorithms are employed to detect the spatial motion information of vehicles. Then, based on perspective transformation, a traffic flow area model is created at the intersection to determine whether traffic events such as overflow or abnormal parking occur. Finally, the reported event information and images are sent to a central server, where a large-scale visual model is used to precisely determine the type of overflow and the cause of abnormal parking. This approach ensures both the real-time nature of the detection task and the accuracy of traffic events, while minimizing the transmission pressure on the traffic intranet and the frequency of large-scale model service calls. Attached Figure Description
[0018] Figure 1 This is a flowchart of the method of the present invention. Figure 2 This is a calibration area diagram of the present invention; Figure 3 This is a perspective transformation diagram of the present invention. Detailed Implementation
[0019] The following is in conjunction with the appendix Figure 1 Appendix Figure 2 Appendix Figure 3 The present invention will be further described below.
[0020] A refined traffic event detection method for intersections based on edge computing perception and multi-source information fusion includes: S1. Obtain the video stream information from the traffic enforcement camera at the intersection, and process the video into frames to obtain... Image set of intersection road images , ,in, For the first Frame intersection road image, .
[0021] S2. In the Frame intersection road image The parameters for determining the overflow area, the straight overflow area, the left turn overflow area, the overflow area, the overflow area, the straight overflow area, and the left turn overflow area are manually calibrated.
[0022] S3. In the Frame intersection road image The four corner points of the perspective transformation region of the quadrilateral are manually marked.
[0023] S4. Obtain the perspective transformation matrix using the four corner points of the perspective transformation region. .
[0024] S5. Regarding the first Frame intersection road image The system performs target detection and tracking on motor vehicles to obtain the target's coordinates, the width of the detection frame, and the height of the detection frame.
[0025] S6. Using the perspective transformation matrix The coordinates of the target, the width of the detection box, and the height of the detection box are calculated to determine the target's position in the first place. The motion rate of the frame.
[0026] S7. Utilize the target in the first Construct a JSON array containing the frame's motion rate, impending overflow region, straight overflow region, left turn overflow region, parameters for determining the impending overflow region, parameters for determining the straight overflow region, and parameters for determining the left turn overflow region. .
[0027] S8. Convert the JSON array Send it to downstream applications or store it in a database.
[0028] By employing a hierarchical processing mechanism that combines rapid edge sensing with in-depth cloud analysis, a balance between efficiency and accuracy in traffic incident detection is achieved, providing reliable technical support for real-time decision-making and intervention in intelligent transportation systems.
[0029] In one embodiment of the present invention, step S1 includes the following steps: S1-1. The edge computing device E-server obtains data from the intersection's electronic police camera via the RTSP protocol. Frame video information.
[0030] S1-2. Using the Ascend Image Processing Unit (DVPP) in C++, the video information stream is decoded frame by frame into BGR format images. The cv::equalizeHist function from the OpenCV library is used to evaluate the first frame. The BGR format image of the frame is enhanced for contrast to obtain the first frame. Frame intersection road image .
[0031] In one embodiment of the present invention, step S2 includes the following steps: S2-1. In the first Frame intersection road image The lower left corner of the quadrilateral's overflow area is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is Construct a distal line parallel to the edge line of the zebra crossing at the front of the opposite exit lane, 20-40 meters along its length. Use the intersection of the left lane line of the opposite exit lane and the distal line as the upper left corner of the quadrilateral's overflow area. The x-coordinate of this point is... The ordinate of this point is The intersection of the right lane line and the far lane line of the opposite exit lane is taken as the upper right corner of the quadrilateral structure's overflow area, and the x-coordinate of this point is... The ordinate of this point is The lower right corner of the quadrilateral's overflow area is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is .
[0032] S2-2. Construct a first proximal line parallel to the edge line of the zebra crossing at the front end of the opposite exit lane, 20-40 meters behind the edge line along its length. Frame intersection road image The lower left corner of the straight overflow area of the quadrilateral structure is defined by the intersection of the extension line of the left lane of the opposite exit lane and the first near-end line. The x-coordinate of this point is: The ordinate of this point is The upper left corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The upper right corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The intersection of the extension line of the right lane of the opposite exit lane and the first near-end line is taken as the lower right corner of the straight overflow area of the quadrilateral structure, and the x-coordinate of this point is . The ordinate of this point is .
[0033] S2-3. At the front end of the zebra crossing of the opposite exit lane, construct a second near-end line parallel to its edge line, 5 meters behind it along its length. At 10-20 meters behind the second near-end line, construct a third near-end line parallel to it. At 5 meters to the left of the left end of the left lane line of the opposite entrance lane, construct a left-end line parallel to it. Frame intersection road image The intersection of the third proximal line and the left end line is taken as the lower left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second proximal line and the left end line is taken as the upper left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second near-end line and the right lane line of the opposite entrance lane is taken as the upper right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the third near-end line and the right lane line of the opposite entrance lane is taken as the lower right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is .
[0034] S2-4. Manually calibrating the parameters for determining the overflow area. , ,in, To determine the threshold for vehicles to linger in the area that is about to overflow, To determine the speed threshold for vehicles to slow down in the area that is about to overflow, To determine the equivalent number of slow-moving standard vehicles in the area about to overflow.
[0035] S2-5. Parameters for manually calibrating the straight-line overflow area , ,in, To determine the threshold for vehicles to linger in the straight-through overflow area, To determine the speed threshold for slow-moving vehicles in the straight-through overflow area, To determine the equivalent number of slow-moving standard vehicles in the straight-through overflow area.
[0036] S2-6. Parameters for manually calibrating the left-turn overflow area , ,in, To determine the threshold for vehicles to linger in the left-turn overflow area, To determine the speed threshold for slow-moving vehicles in the left-turn overflow area, To determine the equivalent number of slow-moving standard vehicles in the left-turn overflow area.
[0037] The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 2.
[0038] In one embodiment of the present invention, step S3 includes the following steps: S3-1. In the... Frame intersection road image The lower left corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite entrance lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is .
[0039] S3-2. Use a GPS locator at the intersection to locate the front edge of the zebra crossing along its length at the front end of the opposite exit lane. Mark a real line parallel to the edge line at that point. The value ranges from 20 to 50 meters, in the... Frame intersection road image The intersection of the left lane line of the opposite entrance lane and the real line is taken as the upper left corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is .
[0040] S3-3. In the... Frame intersection road image The upper right corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is .
[0041] S3-4. No. Frame intersection road image The intersection of the right lane line of the opposite exit lane and the real line is taken as the lower right corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is .
[0042] In one embodiment of the present invention, step S4 uses the function cv::getPerspectiveTransform from the OpenCV library to obtain the coordinates of the four corner points of the perspective transformation region of the quadrilateral structure. , , , ,length Width of the exit lane for the target Convert to a 3x3 perspective transformation matrix .
[0043] In one embodiment of the present invention, step S5 includes the following steps: S5-1. Using the YOLOv11m model to analyze the... Frame intersection road image Target detection is performed on motor vehicles in the middle, and the results are obtained. The nth rectangular target detection box, the nth The x-coordinate of the center point of the detection box of each target is The center point's ordinate is Width is Height is , .
[0044] S5-2. Use the Deepsort tracking algorithm to track each detected target.
[0045] In one embodiment of the present invention, step S6 includes the following steps: S6-1. Through formula Calculation yields the first Information of the detection target in a real-world scenario ,in, For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. These are the perspective transformation coefficients.
[0046] S6-2. Through formula Calculation yields the first The detection target is in the first Motion rate in a frame intersection road image In the formula, For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. This refers to the frame rate of the video from the intersection's electronic police camera, which is parsed by the edge computing device E-server via the RTSP protocol.
[0047] In one embodiment of the present invention, step S7 includes the following steps: S7-1. Using the vector cross product method to utilize the first Coordinates of the center point of the detection bounding box of each target Judge the first Whether the detection box of the first target is in the overflow area, the straight overflow area, or the left turn overflow area, if the first target is in the overflow area. If a detected target is in the overflow region, record the time the target spends in the overflow region. If the first If a detected target is in the straight-line overflow area, record the dwell time of the detected target in the straight-line overflow area. If the first If a detected target is in the left-turn overflow area, record the dwell time of the detected target in the left-turn overflow area. .
[0048] S7-2. If and Then determine the first [item] in the region that is about to overflow. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the impending overflow area among the detected targets is denoted as . .
[0049] S7-3. If and Then determine the first in the straight overflow region. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the straight-through overflow area among the detected targets is denoted as follows: .
[0050] S7-4. If and Then determine the first one in the left turn overflow area. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the left-turn overflow area among the detected targets is denoted as follows: .
[0051] S7-5. Creating Fields ,if and and This indicates normal. ,if and and This indicates that an overflow is imminent. ,if and This indicates straight overflow. ,if and This indicates a left turn overflow. ,if and and This indicates abnormal parking. .
[0052] S7-6. If or or , will the field , , , Store in JSON array middle.
[0053] In one embodiment of the present invention, the JSON array in step S8 Send the data to downstream applications or store it in a database via the Kafka service.
[0054] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A refined traffic event detection method for intersections based on edge computing perception and multi-source information fusion, characterized in that, include: S1. Obtain the video stream information from the traffic enforcement camera at the intersection, and process the video into frames to obtain... Image set of intersection road images , ,in, For the first Frame intersection road image, ; S2. In the Frame intersection road image The parameters for determining the overflow area, the straight overflow area, the left turn overflow area, the overflow area, the overflow area, the straight overflow area, and the left turn overflow area are manually calibrated. S3. In the Frame intersection road image The four corner points of the perspective transformation region of the quadrilateral are manually marked. S4. Obtain the perspective transformation matrix using the four corner points of the perspective transformation region. ; S5. Regarding the first Frame intersection road image The system performs target detection and tracking on motor vehicles to obtain the target's coordinates, the width of the detection box, and the height of the detection box. S6. Using the perspective transformation matrix The coordinates of the target, the width of the detection box, and the height of the detection box are used to calculate the target's position in the first place. Frame motion rate; S7. Utilize the target in the first Construct a JSON array containing the frame's motion rate, impending overflow region, straight overflow region, left turn overflow region, parameters for determining the impending overflow region, parameters for determining the straight overflow region, and parameters for determining the left turn overflow region. ; S8. Convert the JSON array Send to downstream applications or store in a database; Step S3 includes the following steps: S3-1. In the... Frame intersection road image The lower left corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite entrance lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S3-2. Use a GPS locator at the intersection to locate the front edge of the zebra crossing along its length at the front end of the opposite exit lane. Mark a real line parallel to the edge line at that point. The value ranges from 20 to 50 meters, in the... Frame intersection road image The intersection of the left lane line of the opposite entrance lane and the real line is taken as the upper left corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is ; S3-3. In the... Frame intersection road image The upper right corner of the perspective transformation area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S3-4. No. Frame intersection road image The intersection of the right lane line of the opposite exit lane and the real line is taken as the lower right corner of the perspective transformation area of the quadrilateral structure, and the x-coordinate of this point is _____. The ordinate of this point is .
2. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion as described in claim 1, characterized in that, Step S1 includes the following steps: S1-1. The edge computing device E-server obtains data from the intersection's electronic police camera via the RTSP protocol. Frame video information; S1-2. Using the Ascend Image Processing Unit (DVPP) in C++, the video stream is decoded frame by frame into BGR format images. The cv::equalizeHist function from the OpenCV library is used to evaluate the first frame. The BGR format image of the frame is enhanced for contrast to obtain the first frame. Frame intersection road image .
3. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion as described in claim 1, characterized in that, Step S2 includes the following steps: S2-1. In the first Frame intersection road image The lower left corner of the quadrilateral's overflow area is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is [missing information]. The ordinate of this point is Construct a distal line parallel to the edge line of the zebra crossing at the front end of the opposite exit lane, extending 20-40 meters along its length. Use the intersection of the left lane line of the opposite exit lane and the distal line as the upper left corner of the quadrilateral's overflow area. The x-coordinate of this point is... The ordinate of this point is The intersection of the right lane line and the far lane line of the opposite exit lane is taken as the upper right corner of the quadrilateral structure's overflow area, and the x-coordinate of this point is... The ordinate of this point is The lower right corner of the quadrilateral's overflow area is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is ; S2-2. Construct a first proximal line parallel to the edge line of the zebra crossing at the front end of the opposite exit lane, 20-40 meters behind the edge line along its length. Frame intersection road image The lower left corner of the straight overflow area of the quadrilateral structure is defined by the intersection of the extension line of the left lane of the opposite exit lane and the first near-end line. The x-coordinate of this point is: The ordinate of this point is The upper left corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the left lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The upper right corner of the straight overflow area of the quadrilateral structure is defined as the intersection of the right lane line of the opposite exit lane and the front edge of the zebra crossing of the opposite exit lane along its length. The x-coordinate of this point is... The ordinate of this point is The intersection of the extension line of the right lane of the opposite exit lane and the first near-end line is taken as the lower right corner of the straight overflow area of the quadrilateral structure, and the x-coordinate of this point is . The ordinate of this point is ; S2-3. At the front end of the zebra crossing of the opposite exit lane, construct a second near-end line parallel to its edge line, 5 meters behind it along its length. At 10-20 meters behind the second near-end line, construct a third near-end line parallel to it. At 5 meters to the left of the left end of the left lane line of the opposite entrance lane, construct a left-end line parallel to it. Frame intersection road image The intersection of the third proximal line and the left end line is taken as the lower left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second proximal line and the left end line is taken as the upper left corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the second near-end line and the right lane line of the opposite entrance lane is taken as the upper right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is The intersection of the third near-end line and the right lane line of the opposite entrance lane is taken as the lower right corner of the left-turn overflow area of the quadrilateral structure, and the x-coordinate of this point is... The ordinate of this point is ; S2-4. Manually calibrating the parameters for determining the overflow area. , ,in, To determine the threshold for vehicles to linger in the area that is about to overflow, To determine the speed threshold for vehicles to slow down in the area that is about to overflow, To determine the equivalent number of slow-moving standard vehicles in the area about to overflow; S2-5. Parameters for manually calibrating the straight-line overflow area , ,in, To determine the threshold for vehicles to linger in the straight-through overflow area, To determine the speed threshold for slow-moving vehicles in the straight-through overflow area, To determine the equivalent number of slow-moving standard vehicles in the straight-through overflow area; S2-6. Parameters for manually calibrating the left-turn overflow area , ,in, To determine the threshold for vehicles to linger in the left-turn overflow area, To determine the speed threshold for slow-moving vehicles in the left-turn overflow area, To determine the equivalent number of slow-moving standard vehicles in the left-turn overflow area.
4. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion as described in claim 3, characterized in that: The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 4; The value is 10 seconds. The value is taken as 5 km / h. The value is 2.
5. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion as described in claim 1, characterized in that: In step S4, the function cv::getPerspectiveTransform from the OpenCV library is used to obtain the coordinates of the four corner points of the perspective transformation region of the quadrilateral structure. , , , ,length Width of the exit lane for the target Convert to a 3x3 perspective transformation matrix .
6. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion according to claim 1, characterized in that, Step S5 includes the following steps: S5-1. Using the YOLOv11m model to analyze the... Frame intersection road image Target detection is performed on motor vehicles in the middle, and the results are obtained. The nth rectangular target detection box, the nth The x-coordinate of the center point of the detection box of each target is The center point's ordinate is Width is Height is , ; S5-2. Use the Deepsort tracking algorithm to track each detected target.
7. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion according to claim 6, characterized in that, Step S6 includes the following steps: S6-1. Through formula Calculation yields the first Information of the detection target in a real-world scenario ,in, For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. These are the perspective transformation coefficients; S6-2. Through formula Calculation yields the first The detection target is in the first Motion rate in a frame intersection road image In the formula, For the first The detection target is in the first The vertical coordinate in the intersection road image corresponding to the real scene. For the first The detection target is in the first The horizontal coordinate of the intersection road image corresponding to the real scene. This refers to the frame rate of the video from the intersection's electronic police camera, which is parsed by the edge computing device E-server via the RTSP protocol.
8. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion according to claim 3, characterized in that, Step S7 includes the following steps: S7-1. Using the vector cross product method to utilize the first Coordinates of the center point of the detection bounding box of each target Judge the first Whether the detection box of the first target is in the overflow area, the straight overflow area, or the left turn overflow area, if the first target is in the overflow area. If a detected target is in the overflow region, record the time the target spends in the overflow region. If the first If a detected target is in the straight-line overflow area, record the dwell time of the detected target in the straight-line overflow area. If the first If a detected target is in the left-turn overflow area, record the dwell time of the detected target in the left-turn overflow area. ; S7-2. If and Then determine the first element in the region that is about to overflow. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the impending overflow area among the detected targets is denoted as . ; S7-3. If and Then determine the first in the straight overflow region. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the straight-through overflow area among the detected targets is denoted as follows: ; S7-4. If and Then determine the first one in the left turn overflow area. The target of the inspection is slow-moving vehicles. The number of slow-moving vehicles in the left-turn overflow area among the detected targets is denoted as follows: ; S7-5. Creating Fields ,if and and This indicates normal. ,if and and This indicates that an overflow is imminent. ,if and This indicates straight overflow. ,if and This indicates a left turn overflow. ,if and and This indicates abnormal parking. ; S7-6. If or or , field , , , Store in JSON array middle.
9. The method for refined traffic event detection at intersections based on edge computing perception and multi-source information fusion as described in claim 1, characterized in that: JSON array in step S8 Send the data to downstream applications or store it in a database via the Kafka service.
Citation Information
Patent Citations
Traffic signal control method and device, electronic device, and storage medium
CN109410601A
Intersection vehicle overflow parking detection method, device and medium
CN119169822A
