Intelligent detection method for vehicles in tunnel

Abstract

The invention discloses an intelligent detection method for vehicles in a tunnel, and specifically belongs to the technical field of intelligent traffic safety monitoring. The method comprises the following steps: performing target detection and multi-target tracking on a real-time video stream collected by a camera in a tunnel based on an ROI region, and determining a target category attribute list; based on the ROI region, carrying out screening and coordinate conversion on the target category attribute list to obtain an ROI category plane list; performing parking state identification on a target category in the ROI category plane list, determining target vehicles in a parking state, obtaining a first parking list, and calculating a first parking occupancy rate of each lane; normal parking vehicles are removed, and a second parking list is determined; and carrying out accident parking identification on the target category in the second parking list, determining a vehicle in an accident state, and carrying out accident alarm. According to the method, the slight traffic accident can be accurately defined, misjudgment is reduced, the calculation complexity is low, and vehicle accident alarming is carried out in time.

Description

Technical Field

[0001] This application relates to the field of intelligent traffic safety monitoring technology, and in particular to a method for intelligent vehicle detection in tunnels. Background Technology

[0002] In recent years, with the advancement of information technology in the national transportation system, comprehensive monitoring systems have been established for traffic tunnels to monitor the situation inside the tunnels in real time and to provide real-time alarms and handling of traffic incidents. However, manual monitoring is still the primary method of monitoring in tunnels. This manual monitoring method requires relatively high labor costs, and due to the biological characteristics of humans, there may be missed reports when the monitor is not in good condition, thus leading to potential monitoring safety hazards.

[0003] With the development of AI technology, automated monitoring systems based on AI have been gradually deployed, greatly reducing labor costs. However, for minor traffic accidents, existing methods generally suffer from high false alarm rates, such as false alarms caused by construction vehicles and workers, or false alarms due to traffic jams. Furthermore, the inability to clearly define the state of a traffic accident makes current methods trained on datasets unsuitable for detecting minor accidents. In addition, directly analyzing vehicle morphology is computationally complex, requiring extremely high computing resources, and such methods have very low accuracy at distances within camera coverage, making them unsuitable for minor accident detection.

[0004] Therefore, there is an urgent need to develop an intelligent vehicle detection method for tunnels to solve the above problems. Summary of the Invention

[0005] In view of this, this application provides an intelligent vehicle detection method in tunnels, which can accurately identify minor traffic accidents, reduce misjudgments, has low computational complexity, and can promptly issue vehicle accident alarms.

[0006] Specifically, the following technical solutions are included:

[0007] This application provides a method for intelligent vehicle detection in tunnels, the method comprising:

[0008] Step 101: Configure the ROI area of ​​the camera inside the tunnel. The ROI area is the tunnel lane area.

[0009] Step 102: Based on the ROI region, perform target detection on the real-time video stream captured by the camera in the tunnel, determine the target category attribute list, and perform multi-target tracking on the targets in the target category attribute list. The target category attribute list includes the target category, the tracking ID corresponding to the target category, the real-time coordinates of the target's two-dimensional image, and the two-dimensional image coordinates when the target first appears. The target categories include ordinary passenger cars, tunnel engineering vehicles, warning triangles, and ordinary pedestrians.

[0010] Step 103: Based on the ROI region, filter and transform the target category attribute list to obtain the ROI category plane list. The ROI category plane list includes the target category within the ROI region, the tracking ID corresponding to the target category, the real-time plane coordinates of the target, and the plane coordinates of the target when it first appears.

[0011] Step 104: Identify the parking status of the target categories in the ROI category plane list, determine the target vehicles that are parked in the tunnel, obtain the first parking list, and calculate the first parking occupancy rate of each lane.

[0012] Step 105: Remove the normally parked vehicles from the target category in the first parking list to determine the second parking list;

[0013] Step 106: Identify accident parking for the target category in the second parking list, determine the vehicle in an accident state, and issue a vehicle accident alarm.

[0014] In some embodiments, in step 102, target detection is performed using the YOLO series model; and multi-target tracking is performed on the target detection results using bot-sort or bytetrack tracking algorithms.

[0015] In some embodiments, multi-target tracking is performed on targets in the target category attribute list using bot-sort or bytetrack tracking algorithms.

[0016] In some embodiments, step 103 includes:

[0017] Based on the ROI region, the target category attribute list is filtered to determine the ROI category attribute list;

[0018] Perform coordinate transformation on the ROI category attribute list to determine the ROI category plane list.

[0019] In some embodiments, step 104 includes:

[0020] In response to the fact that the change in the coordinate position of a target vehicle in the target category of the ROI category planar list located within the ROI area is less than a distance threshold within a specified time period, it is determined that the target vehicle whose coordinate position change is less than the distance threshold is in a parked state in the tunnel, and the first parking list is obtained.

[0021] The first parking occupancy rate of each lane is obtained by calculating the product of the number of vehicles in each lane and the average vehicle length, as a percentage of the total length of lanes within the coverage area of ​​the cameras in the tunnel.

[0022] In some embodiments, in step 105, normally parked vehicles include construction parking vehicles and vehicles parked due to road congestion.

[0023] In some embodiments, step 105 includes:

[0024] In response to the fact that the occupancy rate of the same lane in the downstream camera in the tunnel is greater than the first occupancy rate threshold, and the target vehicles in the same lane in the current camera in the tunnel are arranged in ascending order of their occupancy, the target vehicles in the corresponding lane in the first parking list located within the range of the camera in the tunnel are determined to be road congestion parking vehicles and are removed.

[0025] In response to the target vehicle in the first parking list being classified as a tunnel engineering vehicle, the target vehicle in the first parking list is determined to be a tunnel engineering vehicle and removed from the list, thus determining the second parking list.

[0026] In some embodiments, step 106 includes:

[0027] If a target vehicle in the second parking list meets the characteristics of an accident vehicle, it is determined to be a vehicle in an accident state, and a vehicle accident alarm is triggered.

[0028] In some embodiments, the characteristics of the accident vehicle include passengers or the driver getting out of the target vehicle, the target vehicle having its hazard lights on, and a warning triangle behind the target vehicle.

[0029] In some embodiments, specific methods for detecting the characteristics of accident vehicles include:

[0030] For passengers or drivers getting off the target vehicle: extract the corresponding data of ordinary pedestrians in the ROI category plane list to obtain the ordinary pedestrian list; traverse the target vehicles in the second parking list, and determine whether the passengers or drivers of the target vehicle get off the vehicle if the number of ordinary pedestrians in the ordinary pedestrian list within the range centered on the target vehicle in the second parking list and with a length threshold as the radius is greater than or equal to one.

[0031] For a target vehicle to activate its hazard lights: Extract the bounding box slice obtained by the real-time video stream captured by the tunnel camera corresponding to the target vehicle in the second parking list and perform target detection. Convert the bounding box slice from RGB channel to HSV channel. If the number of times the element value of the cosine angle between consecutive adjacent vectors in the chromaticity feature value vector change sequence of the HSV channel of the bounding box slice is greater than the angle threshold exceeds the threshold, it is determined that the target vehicle has activated its hazard lights.

[0032] For a target vehicle with a warning triangle behind it: extract the corresponding data of the target category "warning triangle" from the ROI category plane list to obtain the warning triangle list; traverse the target vehicles in the second parking list, and determine that there is a warning triangle behind the target vehicle in response to the presence of a warning triangle behind the target vehicle.

[0033] The beneficial effects of the technical solutions provided in this application include at least the following:

[0034] This application provides an intelligent vehicle detection method for tunnels. The method first determines the Region of Interest (ROI) based on real-time video streams captured by cameras within the tunnel. Then, it performs target detection within the ROI, determines the target detection results, and performs multi-target tracking on these results. The target detection results include target categories and corresponding tracking IDs. Target categories include ordinary passenger cars, tunnel construction vehicles, warning triangles, and ordinary pedestrians. Next, based on the target detection results, it identifies parked vehicles within the tunnel, obtaining a first parking list. Then, it identifies abnormal parking of vehicles in the first parking list, determining a second parking list. Finally, when someone gets out of a vehicle near a target vehicle in the second parking list, a vehicle accident alarm is triggered. This method can accurately identify minor traffic accidents, reduce misjudgments, has low computational complexity, and provides timely vehicle accident alarms. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating the intelligent vehicle detection method in a tunnel provided in an embodiment of this application. Detailed Implementation

[0037] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] To make the technical solutions and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0039] This application provides an intelligent vehicle detection method for tunnels, such as... Figure 1 As shown, the method includes:

[0040] Step 101: Configure the ROI area of ​​the camera inside the tunnel. The ROI area is the tunnel lane area.

[0041] Step 102: Based on the ROI region, perform target detection on the real-time video stream captured by the camera inside the tunnel, determine the target category attribute list, and perform multi-target tracking on the targets in the target category attribute list. The target category attribute list includes the target category, the tracking ID corresponding to the target category, the real-time coordinates of the target's two-dimensional image, and the two-dimensional image coordinates when the target first appears. The target categories include ordinary passenger cars, tunnel engineering vehicles, warning triangles, and ordinary pedestrians.

[0042] In some embodiments, in step 102, target detection is performed using the YOLO series model; and multi-target tracking is performed on the target detection results using bot-sort or bytetrack tracking algorithms.

[0043] In some embodiments, the minimum confidence value of the YOLO series model category is greater than or equal to 0.2. The core parameters of the model inference can be configured as follows: NMS IOU threshold is 0.7, half-precision inference (half) is set to True, and video frame stride (vid_stride) is set to 1. For bot-sort or bytetrack tracking algorithms, the confidence threshold for the second matching stage is 0.55, the confidence threshold for the first matching stage is 0.1, the confidence threshold for generating a new ID is 0.5, the tracking buffer window size is 24 frames, and the pairing computation complexity threshold is 0.9.

[0044] Step 103: Based on the ROI region, filter and transform the target category attribute list to obtain the ROI category plane list. The ROI category plane list includes the target category within the ROI region, the tracking ID corresponding to the target category, the real-time plane coordinates of the target, and the plane coordinates of the target when it first appears.

[0045] In some embodiments, step 103 includes:

[0046] Step 1031: Based on the ROI region, filter the target category attribute list to determine the ROI category attribute list.

[0047] Step 1032: Perform coordinate transformation on the ROI category attribute list to determine the ROI category plane list. Convert the image coordinates of the target category to plane coordinates to obtain the coordinate position of the target category in the plane coordinate system, thereby determining the position of the target category in the tunnel.

[0048] In some embodiments, the midpoint of the lower edge of the target detection bounding box is marked as the coordinate position of the target vehicle in the image coordinate system.

[0049] In some embodiments, the transformation matrix MAT is obtained through prior calibration, and then the target category's coordinate position in the planar coordinate system is obtained by projecting the image coordinates into homogeneous coordinates, i.e., the homogeneous coordinates of the target category on the road surface.

[0050] Step 104: Identify the parking status of the target categories in the ROI category plane list, determine the target vehicles that are parked in the tunnel, obtain the first parking list, and calculate the first parking occupancy rate of each lane.

[0051] In some embodiments, step 104 includes:

[0052] Step 1041: In response to the fact that the change in the coordinate position of a target vehicle in the target category of the ROI category planar list located within the ROI area is less than a distance threshold within a specified time period, it is determined that the target vehicle whose coordinate position change is less than the distance threshold is in a parked state in the tunnel, and a first parked list is obtained; by detecting the position change of the target vehicle in the tunnel within a specified time period, it can be determined whether the target vehicle is in a parked state.

[0053] If the change in the coordinate position of a target vehicle in the target category of the ROI category list within the ROI area is less than the distance threshold within a specified time period, it indicates that the target vehicle may not have moved and is a suspected accident vehicle.

[0054] In some embodiments, the specified duration can be 3 seconds.

[0055] In some embodiments, the distance threshold can be 3m.

[0056] Step 1042: Calculate the ratio of the product of the number of vehicles in each lane and the average vehicle length to the total length of lanes within the coverage area of ​​the cameras in the tunnel, obtaining the first parking occupancy rate for each lane. The first parking occupancy rate reflects the road congestion situation. Determining this state value can prepare for subsequent judgments as to whether a small number of vehicles in a certain upstream camera area are passively congested due to congestion in its downstream camera area.

[0057] In some embodiments, the specific formula for implementing step 1042 can be:

[0058]

[0059] Where Oi is the first parking occupancy rate of the i-th lane, n is the number of vehicles in the i-th lane, K is the average vehicle length, which can be 6 meters (vehicle length + vehicle spacing), and L is the radiation range of the camera in the tunnel (i.e., the total length of the lane).

[0060] Step 105: Remove normal parking vehicles from the target category in the first parking list to determine the second parking list, thereby avoiding misclassifying construction parking vehicles and road congestion parking vehicles as accident vehicles.

[0061] In some embodiments, in step 105, normally parked vehicles include construction parking vehicles and vehicles parked due to road congestion.

[0062] In some embodiments, step 105 includes:

[0063] Step 1051: In response to the fact that the occupancy rate of the same lane in the downstream camera in the tunnel is greater than the first occupancy rate threshold, and the target vehicles in the same lane in the current camera in the tunnel are arranged in ascending order of their occupancy, the target vehicles in the corresponding lanes within the range of the cameras in the tunnel in the first parking list are determined to be road congestion parking vehicles and are removed from the list, thereby avoiding misjudging road congestion parking vehicles as accident vehicles.

[0064] It should be noted that the occupancy rate of the same lane in the downstream cameras inside the tunnel is greater than the first occupancy rate threshold, indicating that there are many vehicles in the downstream of the tunnel, and the vehicles occupy a large proportion of the lane, which may cause traffic jams. Further investigation is needed to eliminate parked vehicles that are causing road congestion.

[0065] In some embodiments, the first occupancy threshold may be 0.9.

[0066] Step 1052: In response to the target category of the target vehicle in the first parking list being a tunnel construction vehicle, the target vehicle in the first parking list with the target category of tunnel construction vehicle is determined to be a tunnel construction vehicle and removed from the list, and the second parking list is determined, thereby avoiding misjudging construction parking vehicles as accident vehicles.

[0067] In real-world traffic scenarios, parking incidents caused by morning and evening rush hours and construction vehicles account for more than 95% of all parking incidents. Step 1052 greatly reduces the false alarm rate of the system by eliminating these two situations, reduces the workload of manual secondary review, and increases the workload of traffic management personnel.

[0068] Step 106: Identify accident parking for the target category in the second parking list, determine the vehicle in an accident state, and issue a vehicle accident alarm.

[0069] In some embodiments, step 106 includes:

[0070] If a target vehicle in the second parking list meets the characteristics of an accident vehicle, it is determined to be a vehicle in an accident state, and a vehicle accident alarm is triggered.

[0071] In some embodiments, the characteristics of the accident vehicle include passengers or the driver getting out of the target vehicle, the target vehicle having its hazard lights on, and a warning triangle behind the target vehicle.

[0072] In some embodiments, specific methods for detecting the characteristics of accident vehicles include:

[0073] (1) For passengers or drivers getting off the target vehicle: Extract the corresponding data of ordinary pedestrians in the ROI category plane list to obtain the ordinary pedestrian list; traverse the target vehicles in the second parking list. If the number of ordinary pedestrians in the ordinary pedestrian list within the range centered on the target vehicle in the second parking list and with a length threshold as the radius is greater than or equal to one, it is determined that the passenger or driver got off the target vehicle. The corresponding data of ordinary pedestrians in the ROI category plane list can reflect the position of the ordinary pedestrian when it first appears. Combined with the fact that the number of ordinary pedestrians in the ordinary pedestrian list within the range centered on the target vehicle in the second parking list and with a length threshold as the radius is greater than or equal to one, it indicates that the pedestrian got off the target vehicle and is not an outsider (such as a construction worker, police officer, or onlooker). This completes the determination that there are people near the target vehicle and avoids misjudging outsiders.

[0074] (2) For the target vehicle to turn on the hazard warning lights: extract the bounding box slice obtained by the real-time video stream collected by the tunnel camera corresponding to the target vehicle in the second parking list and perform target detection. Convert the bounding box slice from RGB channel to HSV channel. If the number of times the element value of the cosine angle between consecutive adjacent vectors in the chromaticity feature value vector change sequence of the HSV channel of the bounding box slice is greater than the angle threshold exceeds the number threshold, it is determined that the target vehicle turns on the hazard warning lights.

[0075] In some embodiments, in response to the number of times the element value of the cosine angle value sequence of consecutive adjacent vectors in the chromaticity feature value vector change sequence of the HSV channel of the bounding box slice is continuously greater than the angle threshold, determining that the target vehicle should activate its hazard warning lights may specifically include:

[0076] After converting each pixel of the HSV channel bounding box slice into a normalized three-dimensional vector, the mean of each dimension is calculated to obtain the chromaticity feature vector V_car = (H_avg, S_avg, V_avg) representing the bounding box slice of the target vehicle, where H_avg, S_avg, and V_avg are the normalized average hue, average saturation, and average brightness, respectively.

[0077] The chromaticity feature vector V_car is continuously sampled in time, with the sampling frequency consistent with the hazard light frequency of a typical vehicle, to obtain the chromaticity feature vector change sequence V_k_list = [V_1, V_2, ......V_n], where V_n is the vector value of the k-th vehicle at the nth sampling time.

[0078] Calculate the directional similarity of consecutive adjacent vectors in the chromaticity feature value vector change sequence V_k_list, that is, calculate the cosine angle value A of consecutive adjacent vectors, and obtain the cosine angle value sequence A = [A_1,A_2,......,A_n-1].

[0079] The number of times an element in the cosine angle sequence A is continuously greater than the angle threshold is counted. If the number of times exceeds the threshold, the target vehicle is determined to activate its hazard lights.

[0080] (3) For vehicles with a warning triangle behind them: extract the corresponding data of vehicles with the warning triangle category from the ROI category plane list to obtain the warning triangle list; traverse the target vehicles in the second parking list, and determine that there is a warning triangle behind the target vehicle in response to the presence of a warning triangle behind the target vehicle.

[0081] In real-world tunnel traffic scenarios, drivers involved in accidents typically perform the following sequential actions: activating hazard lights, getting out of the vehicle to inspect the damage, and setting up a warning triangle. Therefore, by simultaneously identifying these different types of traffic accident signals, the detection rate of real-world events can be improved, significantly reducing the possibility of missing reports of actual traffic incidents due to the absence of a single feature.

[0082] In summary, the embodiments of this application disclose an intelligent vehicle detection method in tunnels. This method can accurately identify minor traffic accidents, reduce misjudgments, has low computational complexity, and can promptly issue vehicle accident alarms.

[0083] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "multiple" refers to two or more unless otherwise expressly defined.

[0084] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only.

[0085] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for detecting a vehicle in a tunnel, characterized in that, The method includes: Step 101: Configure the ROI area of ​​the camera inside the tunnel. The ROI area is the tunnel lane area. Step 102: Based on the ROI region, perform target detection on the real-time video stream captured by the camera in the tunnel, determine the target category attribute list, and perform multi-target tracking on the targets in the target category attribute list. The target category attribute list includes the target category, the tracking ID corresponding to the target category, the real-time coordinates of the target's two-dimensional image, and the two-dimensional image coordinates when the target first appears. The target categories include ordinary passenger cars, tunnel engineering vehicles, warning triangles, and ordinary pedestrians. Step 103: Based on the ROI region, filter and transform the target category attribute list to obtain the ROI category plane list. The ROI category plane list includes the target category within the ROI region, the tracking ID corresponding to the target category, the real-time plane coordinates of the target, and the plane coordinates of the target when it first appears. Step 104: Identify the parking status of the target categories in the ROI category plane list, determine the target vehicles that are parked in the tunnel, obtain the first parking list, and calculate the first parking occupancy rate of each lane. Step 105: Remove the normally parked vehicles from the target category in the first parking list to determine the second parking list; Step 106: Identify accident parking for the target category in the second parking list, determine the vehicle in an accident state, and issue a vehicle accident alarm.

2. The method of claim 1, wherein, In step 102, target detection is performed using the YOLO series models; multi-target tracking is performed on the target detection results using bot-sort or bytetrack tracking algorithms.

3. The method of claim 1, wherein, Multi-target tracking is performed on targets in the target category attribute list using bot-sort or bytetrack tracking algorithms.

4. The method of claim 3, wherein, Step 103 includes: Based on the ROI region, the target category attribute list is filtered to determine the ROI category attribute list; Perform coordinate transformation on the ROI category attribute list to determine the ROI category plane list.

5. The method of claim 1, wherein, Step 104 includes: In response to the fact that the change in the coordinate position of a target vehicle in the target category of the ROI category planar list located within the ROI area is less than a distance threshold within a specified time period, it is determined that the target vehicle whose coordinate position change is less than the distance threshold is in a parked state in the tunnel, and the first parking list is obtained. The first parking occupancy rate of each lane is obtained by calculating the product of the number of vehicles in each lane and the average vehicle length, as a percentage of the total length of lanes within the coverage area of ​​the cameras in the tunnel.

6. The method of claim 1, wherein, In step 105, normally parked vehicles include construction vehicles and vehicles parked due to road congestion.

7. The method of claim 6, wherein, Step 105 includes: In response to the fact that the occupancy rate of the same lane in the downstream camera in the tunnel is greater than the first occupancy rate threshold, and the target vehicles in the same lane in the current camera in the tunnel are arranged in ascending order of their occupancy, the target vehicles in the corresponding lane in the first parking list located within the range of the camera in the tunnel are determined to be road congestion parking vehicles and are removed. In response to the target vehicle in the first parking list being classified as a tunnel engineering vehicle, the target vehicle in the first parking list is determined to be a tunnel engineering vehicle and removed from the list, thus determining the second parking list.

8. The method of claim 1, wherein, Step 106 includes: If a target vehicle in the second parking list meets the characteristics of an accident vehicle, it is determined to be a vehicle in an accident state, and a vehicle accident alarm is triggered.

9. The method of claim 8, wherein, The characteristics of the accident vehicle include passengers or the driver getting out of the vehicle, the vehicle having its hazard lights on, and a warning triangle behind the vehicle.

10. The method of claim 9, wherein, Specific methods for detecting the characteristics of accident vehicles include: For passengers or drivers getting off the target vehicle: extract the corresponding data of ordinary pedestrians in the ROI category plane list to obtain the ordinary pedestrian list; traverse the target vehicles in the second parking list, and determine whether the passengers or drivers of the target vehicle get off the vehicle if the number of ordinary pedestrians in the ordinary pedestrian list within the range centered on the target vehicle in the second parking list and with a length threshold as the radius is greater than or equal to one. For a target vehicle to activate its hazard lights: Extract the bounding box slice obtained by the real-time video stream captured by the tunnel camera corresponding to the target vehicle in the second parking list and perform target detection. Convert the bounding box slice from RGB channel to HSV channel. If the number of times the element value of the cosine angle between consecutive adjacent vectors in the chromaticity feature value vector change sequence of the HSV channel of the bounding box slice is greater than the angle threshold exceeds the threshold, it is determined that the target vehicle has activated its hazard lights. For a target vehicle with a warning triangle behind it: extract the corresponding data of the target category "warning triangle" from the ROI category plane list to obtain the warning triangle list; traverse the target vehicles in the second parking list, and determine that there is a warning triangle behind the target vehicle in response to the presence of a warning triangle behind the target vehicle.

Images