Zebra crossing pedestrian-vehicle collision early warning method and system
Video data is collected through cameras, pedestrians and vehicles are identified using YOLOv5 and monocular ranging algorithm, human-vehicle distances are calculated and collision risks are evaluated, and drones are dispatched to provide early warnings, which solves the early warning problem of human-vehicle collision accidents on roads without signal lights, and realizes the The effect of reducing accidents.
Patent Information
- Application Number
- CN202410883253.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-03
- Publication Date
- 2024-11-12
AI Technical Summary
On road sections without signal lights, pedestrians are blocked from sight when crossing zebra crossing, which makes it difficult for drivers to respond in a timely manner, resulting in human-vehicle collision accidents. The existing technology lacks effective early warning plans.
By setting up a camera to collect video data, using the YOLOv5 algorithm to identify pedestrians and vehicles, combining a monocular distance measurement algorithm to calculate the distance between people and vehicles, establish a minimum safety distance model and risk assessment model. If it is determined that the collision probability is greater than the threshold, send a drone to conduct early warning .
Timely and accurate warnings for human-vehicle collisions have been achieved to reduce traffic accidents.
Smart Images

Figure CN118942277A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to traffic safety monitoring, and in particular to a zebra crossing human-vehicle collision early warning method and system. Background Art
[0002] When pedestrians cross zebra crossings, especially on roads without traffic lights, pedestrians suddenly appear from the driver's blind spot due to obstructed vision (i.e. "human-vehicle collision"), making it difficult for the driver to react in time, thus causing traffic accidents. Early warning of "human-vehicle collision" and timely reminders to pedestrians and drivers can help reduce the occurrence of traffic accidents. At present, there is a lack of effective and feasible "human-vehicle collision" early warning solutions. Summary of the invention
[0003] Purpose of the invention: The first purpose of the present invention is to provide a zebra crossing human-vehicle collision warning method with high accuracy and feasibility; the second purpose of the present invention is to provide a zebra crossing human-vehicle collision warning system.
[0004] Technical solution: The zebra crossing vehicle collision warning method of the present invention comprises:
[0005] (1) The video data in the set monitoring area is collected by a camera installed on one side of the sidewalk, and image recognition is used to determine whether there are pedestrians crossing the street on the other side of the sidewalk and whether the rear vehicle's field of view is blocked by the front vehicle or other objects. If both conditions are met, step (2) is executed;
[0006] (2) Calculate the distance L between the rear vehicle and the camera using a monocular ranging algorithm 1 and the distance L between the pedestrian and the camera 2 , determine the distance S between the person and the vehicle;
[0007] (3) Establish a minimum safe distance model between people and vehicles and determine the longitudinal safe distance D x , when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When , execute step (4); γ is the angle between the straight line where the pedestrian and the vehicle are located and the direction of vehicle travel;
[0008] (4) Establish a risk assessment model to determine the probability of a "human-vehicle collision". If the probability is greater than or equal to a threshold value P, execute step (5); otherwise, it is considered that the "human-vehicle collision" will not occur.
[0009] (5) Two drones are used for early warning, one is deployed above the sidewalk to alert pedestrians, and the other is deployed on the side of the road to alert drivers.
[0010] Furthermore, in step (1), image recognition is performed using the YOLOv5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street. The YOLOv5 algorithm is used to detect vehicles and generate a bounding box for each vehicle. The coordinates of the bounding boxes are used to analyze the positions of these bounding boxes and the degree of overlap, and the spatial relationship between the vehicles is determined. If the bounding box of one vehicle is above or in front of the bounding box of another vehicle, and the bounding boxes of the two vehicles overlap, the field of view of the rear vehicle is blocked by the front vehicle.
[0011] Furthermore, in step (2), the distance L between the rear vehicle and the camera is 1 and the distance L between the pedestrian and the camera 2 The calculation formula is as follows:
[0012]
[0013] Where i = 1, 2; C is a constant; BP is the size of the actual object in the picture in the Bpix camera;
[0014] Connect the camera, pedestrians and vehicles to form a triangle, and calculate the distance S between the pedestrian and the vehicle based on the triangle relationship:
[0015]
[0016] Among them, ω is the angle between the vehicle and pedestrian and the camera.
[0017] Further,
[0018]
[0019] Among them, R is the radius of the vehicle or pedestrian to be tested; AP is the Apix fixed pixel, that is, the pixel size; α is the field of view angle;
[0020] The pixel size of the reference object in the image is measured using the YOLOv5 algorithm, and a scaling factor is obtained based on the pixel size and the known actual size of the reference object. This scaling factor is used to convert the pixel size of the vehicle or pedestrian in the image into the actual size.
[0021] First, determine the bounding box of the reference object in the image and measure the pixel value of its width; let the width of the reference object in the image be P wref pixels, the actual width of the reference object is known, set to W ref ; Use the pixel size of the reference object and the actual size to calculate the width scale factor S w :
[0022]
[0023] For the vehicle or pedestrian to be measured in the image, the pixel width P of its bounding box is also measuredwobj , the pixel size of the vehicle or pedestrian is converted to the actual size through the above-calculated scaling factor; the actual width of the vehicle or pedestrian W obj Calculated by the following formula:
[0024] W obj =P wobj ×S w
[0025] The vehicle or pedestrian radius R passes through the actual width W obj Divide by 2 to get:
[0026]
[0027] Furthermore, in step (3), the minimum safe distance model between people and vehicles does not consider the lateral offset, that is, the offset perpendicular to the road direction. The longitudinal safety distance is the minimum safe distance between people and vehicles along the road direction, that is, the longitudinal distance D is calculated. x ;
[0028]
[0029] Among them, v v represents the initial speed of the vehicle before braking, t s represents the driver's reaction time, a max Indicates the maximum braking deceleration during braking, v p Indicates the speed of pedestrians when the vehicle brakes, t p It indicates the time for pedestrians to walk after the vehicle decelerates, through v p t p cosθ can be used to calculate the distance traveled longitudinally by the pedestrian, where θ is the heading angle of the vehicle; d s It indicates the safe parking distance, that is, the acceptable safe parking distance between cars and pedestrians.
[0030] Furthermore, in step (4), a risk assessment model is established using environmental impact parameters to determine the potential risk P(γ sk=1 ), environmental impact parameters include drivable channel c r , Traffic volume under obstruction of visionω 0 , number of lanes crossed n 1 , speed of the obstruction v 0 , Isolation belt d i , headway time t h and spatiotemporal contexts t ;
[0031] Driving channel c r The value is assigned by judging whether the lane where the "vehicle-person collision" phenomenon occurs is a lane that traffic participants can drive in accordance with traffic regulations:
[0032]
[0033] Traffic volume under obstructed visionω 0 By assigning the flow level of traffic participants around the blocked area:
[0034]
[0035] Number of lanes crossed n 1 The value is assigned by the number of one-way lanes that pedestrians cross after the visual field is blocked:
[0036]
[0037] Obstruction speed v 0 is assigned by the speed level of the occluder moving toward the occluded area, where the occlusion speed v of the static occluder is 0 =0:
[0038]
[0039] Isolation zone i The value is assigned based on whether there are isolation barriers on the passageway that the traffic participants cross:
[0040]
[0041] Headway time t h The value is assigned based on the headway between the vehicle in the "vehicle-person collision" lane and the vehicle in front of it in the same lane:
[0042]
[0043] Spatiotemporal contexts t By considering rush hour or places where pedestrians are likely to appear, such as schools and residential areas, the value is assigned as:
[0044]
[0045] Potential risks sk By comprehensively considering the seven environmental impact factors, they are classified into two situations: safety and risk, namely:
[0046]
[0047] Assume that the average number of people entering the visual field obstruction area per minute is ω. Since the behavior of pedestrians entering the visual field obstruction area is random, the number of pedestrians entering the obstruction area obeys the Poisson distribution. Then the probability of a "human-vehicle collision" when pedestrians randomly cross the road is:
[0048]
[0049] Among them, P(γ sk =1) indicates the probability of a "pedestrian-vehicle collision" when a pedestrian crosses the road; 1-e -ω / 60 It represents the probability of pedestrians reaching the visual field obstruction area per unit time; e -ω / 60 =P(N(t=1s))=0 is the probability that no one reaches the blocked area; K ρ is the environmental impact factor.
[0050] Further,
[0051]
[0052] Among them, k c represents the “probability coefficient of traffic participants crossing the road illegally” and follows a normal distribution; k d represents the “isolation zone influence coefficient”; k t represents the “headway influence coefficient”; k v Indicates "Obstacle Speed Influence Coefficient".
[0053] Furthermore, k c ~N(0.4,0.02), k d =0.46, k v =1.45, k t =0.05.
[0054] Furthermore, the threshold value P = 0.35 is configured, and P(γ sk =1) and the threshold value P for early warning:
[0055]
[0056] The zebra crossing vehicle collision warning system of the present invention comprises:
[0057] Image acquisition module, used to collect video data of vehicles and pedestrians in the set monitoring area;
[0058] The image recognition module is used to perform image recognition on the collected vehicle and pedestrian video data using the YOLOv5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street. At the same time, it identifies whether there is a vehicle and determines whether the rear vehicle's field of view is blocked. If there is a field of view blockage, the distance L between the vehicle and the camera is calculated using a monocular ranging algorithm. 1 and the distance L between the pedestrian and the camera 2 , calculate the distance S between the person and the vehicle, and determine whether the longitudinal component Scosγ of the distance S is less than or equal to the longitudinal safety distance D x ;
[0059] Risk assessment module, used when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When the vehicle is in the vehicle, the probability of a "human-vehicle collision" is determined according to the risk assessment model;
[0060] And, the early warning module is used to dispatch two drones for early warning when the probability of a "human-vehicle collision" is greater than or equal to a threshold P. One drone is deployed above the sidewalk to remind pedestrians, and the other is deployed on the side of the road to remind the driver.
[0061] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: By combining the YOLOv5 algorithm with the monocular ranging algorithm, the present invention can timely and accurately detect the field of view obstruction problem between vehicles and judge whether there is a risk of conflict and collision between people and vehicles. When there is a risk, the drone can be used to issue a timely warning to reduce the occurrence of "people-vehicle collision" traffic accidents. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 is a flow chart of a zebra crossing vehicle collision warning method provided by an embodiment of the present invention;
[0063] Figure 2 This is a schematic diagram of the monocular distance measurement principle in an embodiment of the present invention;
[0064] Figure 3 Schematic diagram of the structure of the zebra crossing vehicle collision warning system in an embodiment of the present invention;
[0065] Figure 4 This is an application scenario of the zebra crossing human-vehicle collision warning method and system in an embodiment of the present invention. DETAILED DESCRIPTION
[0066] The present invention will be further described below in conjunction with the accompanying drawings.
[0067] like Figure 1 and Figure 4 As shown, the embodiment of the present invention provides a zebra crossing vehicle collision warning method, which specifically includes the following steps:
[0068] (1) A camera installed on one side of the sidewalk collects video data within a set monitoring area, and through image recognition, determines whether there are pedestrians crossing the street on the other side of the sidewalk and whether the rear vehicle's field of view is blocked by the front vehicle or other objects (such as green belts). If both conditions are met, execute step (2);
[0069] In step (1), image recognition is performed using the YOLO v5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street.
[0070] The YOLOv5 algorithm is used to detect vehicles and generate bounding boxes for each vehicle. The coordinates of the bounding boxes are used to analyze the positions and overlaps of these bounding boxes to determine the spatial relationship between vehicles. If the bounding box of one vehicle is above or in front of the bounding box of another vehicle, and the bounding boxes of the two vehicles overlap, it means that the view of the rear vehicle is blocked by the front vehicle.
[0071] (2) Calculate the distance L between the rear vehicle and the camera using a monocular ranging algorithm 1 and the distance L between the pedestrian and the camera 2 , determine the distance S between the person and the vehicle;
[0072] In step (2), the pixel size of the reference object in the image is measured by the YOLOv5 algorithm, and a scaling factor is obtained based on the pixel size and the known actual size of the reference object. This scaling factor is used to convert the pixel size of the vehicle or pedestrian in the image into the actual size. In this embodiment, the wheel diameter of a standard car is selected as the reference object.
[0073] First, determine the bounding box of the reference object in the image and measure its width in pixels. Let the width of the reference object in the image be P wref pixels, the actual width of the reference object is known, set to W ref Using the pixel size of the reference object and the actual size, calculate the width scale factor S w :
[0074]
[0075] For the vehicle or pedestrian to be measured in the image, the pixel width P of its bounding box is also measured wobj , the pixel size of the vehicle or pedestrian is converted to the actual size by the above-calculated scaling factor. The actual width of the vehicle or pedestrian W obj Calculated by the following formula:
[0076] W obj =P wobj ×S w
[0077] The vehicle or pedestrian radius R passes through the actual width W ooj Divide by 2 to get:
[0078]
[0079] The monocular distance measurement algorithm calculates the object through the optical center of the lens and maps it onto the image sensor (image plane). An inverted image will appear on the image plane. By measuring the actual object, the distance L between the vehicle in the lane and the camera is obtained. 1 The distance L between the pedestrian on the zebra crossing and the camera 2 . Combined Figure 2 The schematic diagram shown, specifically,
[0080]
[0081] Among them, α is the field of view angle; AP is the Apix fixed pixel, that is, the pixel size; BP is the size of the actual object in the picture in the Bpix camera; L' is the focal length of the camera; β is the angle between the object and the optical center of the lens; R is the radius of the vehicle or pedestrian to be tested; H is the width;
[0082] Combining the above formulas, we can get:
[0083]
[0084] Where i = 1, 2; is a constant, let it be C, we can get:
[0085]
[0086] Connect the camera, pedestrians and vehicles to form a triangle, and calculate the distance S between the pedestrian and the vehicle based on the triangle relationship:
[0087]
[0088] Among them, L 1 is the distance between the vehicle in the lane and the camera, L 2 is the distance between the pedestrian on the zebra crossing and the camera, and ω is the angle between the vehicle and the pedestrian and the camera.
[0089] (3) Establish a minimum safe distance model between people and vehicles and determine the longitudinal safe distance D x , when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When , execute step (4); γ is the angle between the straight line where the pedestrian and the vehicle are located and the direction of vehicle travel;
[0090] In step (3), the minimum safe distance model between pedestrians and vehicles does not consider the lateral deviation caused by the vehicle's failure to detect pedestrians in time during braking. Under ideal conditions, the drone can warn the vehicle in the "vehicle-pedestrian collision" lane before it detects pedestrians, thereby achieving the purpose of deceleration. During this period, no lateral deviation will occur. Therefore, the lateral deviation, that is, the deviation perpendicular to the road direction, is not considered. The longitudinal safety distance is the minimum safe distance between the vehicle and the pedestrian along the road direction, that is, the longitudinal distance D is calculated. x ;
[0091]
[0092] Among them, v vIndicates the initial speed of the vehicle before braking; t s Indicates the driver's reaction time, which is about 0.2s; a max Indicates the maximum braking deceleration during braking, v p Indicates the speed of pedestrians when the vehicle brakes, t p It indicates the time for pedestrians to walk after the vehicle decelerates, through v p t p cosθ can be used to calculate the distance traveled longitudinally by the pedestrian, where θ is the heading angle of the vehicle; d s It indicates the safe parking distance, that is, the acceptable safe parking distance between cars and pedestrians.
[0093] (4) Establish a risk assessment model to determine the probability of a "human-vehicle collision". If the probability is greater than or equal to a threshold value P, execute step (5); otherwise, it is considered that the "human-vehicle collision" will not occur.
[0094] In step (4), the environmental impact parameters are used to establish a risk assessment model to determine the potential risk P(γ sk=1 ), environmental impact parameters include drivable channel c r , Traffic flow under obstructed visionω 0 , number of lanes crossed n 1 , speed of the obstruction v 0 , Isolation belt d i , headway time t h and spatiotemporal contexts t ;
[0095] Driving channel c t The value is assigned by judging whether the lane where the "vehicle-person collision" phenomenon occurs is a lane that traffic participants can drive in accordance with traffic regulations:
[0096]
[0097] Traffic flow under obstructed visionω 0 By assigning the flow level of traffic participants around the blocked area:
[0098]
[0099] Number of lanes crossed n 1 The value is assigned by the number of one-way lanes that pedestrians cross after the visual field is blocked:
[0100]
[0101] Obstruction speed v 0 is assigned by the speed level of the occluder moving toward the occluded area, where the occlusion speed v of the static occluder is 0 =0:
[0102]
[0103] Isolation zone i The value is assigned based on whether there are isolation barriers on the passageway that the traffic participants cross:
[0104]
[0105] Headway time t h The value is assigned based on the headway between the vehicle in the "vehicle-person collision" lane and the vehicle in front of it in the same lane:
[0106]
[0107] Spatiotemporal contexts t By considering rush hour or places where pedestrians are likely to appear, such as schools and residential areas, the value is assigned as:
[0108]
[0109] Potential risks sk By comprehensively considering the seven environmental impact factors, they are classified into two situations: safety and risk, namely:
[0110]
[0111] Assume that the average number of people entering the visual field obstruction area per minute is ω. Since the behavior of pedestrians entering the visual field obstruction area is random, the number of pedestrians entering the obstruction area obeys the Poisson distribution. Then the probability of a "human-vehicle collision" when pedestrians randomly cross the road is:
[0112]
[0113] Among them, P(γ dk =1) indicates the probability of a "pedestrian-vehicle collision" when a pedestrian crosses the road; 1-e -ω / 60 represents the probability of pedestrians reaching the visual field obstruction area per unit time; e -ω / 60 =P(N(t=1s))=0 is the probability that no one reaches the blocked area; K ρ is the environmental impact coefficient, which represents the influence of the selected environmental factors on the event of “pedestrians rushing out of the blocked area”.
[0114]
[0115] Among them, k c represents the “probability coefficient of traffic participants crossing the road illegally” and follows a normal distribution; k d represents the “isolation zone influence coefficient”; k yrepresents the “headway influence coefficient”; k b represents the "obstacle speed influence coefficient". It can be seen that K ρ By drivable channel c t , number of lanes crossed n 1 , speed of the obstruction v 0 , Isolation belt d o , headway time t h and spatiotemporal contexts t These six environmental impact parameters jointly determine the probability of a "pedestrian-vehicle collision" incident when pedestrians cross the road under different traffic conditions.
[0116] (5) Two drones are used for early warning, one is deployed above the sidewalk to alert pedestrians, and the other is deployed on the side of the road to alert drivers.
[0117] Configure the threshold P = 0.35, compare P(γ sk =1) and the threshold value P for early warning:
[0118]
[0119] A specific example is given below.
[0120] A road section without traffic lights is selected. The road section is a two-way four-lane road section without a central dividing strip and with dense traffic and easy presence of pedestrians.
[0121] According to the characteristics of the road section, the following values are assigned:
[0122] Spatiotemporal contexts t =1, drivable lane c r =1, isolation zone d i = 0, number of lanes crossed n 1 = 4. This example is aimed at studying stationary obstructions, so let the obstruction speed v 0 = 0, where the headway time between vehicles in the “vehicle-person collision” lane is kept within 2.7 / veh, and the headway time between vehicles is set to t h =0, by different flow rates ω 0 The situation is calculated to determine the probability of "human-vehicle collision" at each flow level.
[0123] According to the value of each coefficient k c ~N(0.4,0.02), k d =0.46, k v =1.45, k t =0.05, and we get K ρ =0.68.
[0124] Kρ =0.68 into According to different flow levels ω 0 Assign values to obtain the probability of "human-vehicle collision" under different traffic levels, that is:
[0125]
[0126] The probability of a "human-vehicle collision" is compared with the set threshold of 0.35, and a drone is used for early warning.
[0127] like Figure 3 As shown, the embodiment of the present invention also provides a zebra crossing vehicle collision warning system, including:
[0128] Image acquisition module, used to collect video data of vehicles and pedestrians in the set monitoring area;
[0129] The image recognition module is used to perform image recognition on the collected vehicle and pedestrian video data using the YOLOv5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street. At the same time, it identifies whether there is a vehicle and determines whether the rear vehicle's field of view is blocked. If there is a field of view blockage, the distance L between the vehicle and the camera is calculated using a monocular ranging algorithm. 1 and the distance L between the pedestrian and the camera 2 , calculate the distance S between the person and the vehicle, and determine whether the longitudinal component Scosγ of the distance S is less than or equal to the longitudinal safety distance D x ;
[0130] Risk assessment module, used when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When the vehicle is in the vehicle, the probability of a "human-vehicle collision" is determined according to the risk assessment model;
[0131] And, the early warning module is used to dispatch two drones for sound and light warning when the probability of a "human-vehicle collision" is greater than or equal to a threshold P. One drone is deployed above the sidewalk to remind pedestrians, and the other is deployed on the side of the road to remind the driver.
Claims
1. A zebra crossing vehicle collision warning method, characterized in that: include: (1) The video data in the set monitoring area is collected by a camera installed on one side of the sidewalk, and image recognition is used to determine whether there are pedestrians crossing the street on the other side of the sidewalk and whether the rear vehicle's field of view is blocked by the front vehicle or other objects. If both conditions are met, step (2) is executed; (2) Using a monocular ranging algorithm, calculate the distance L1 between the rear vehicle and the camera and the distance L2 between the pedestrian and the camera, and determine the distance S between the pedestrian and the vehicle; (3) Establish a minimum safe distance model between people and vehicles and determine the longitudinal safe distance D x , when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When , execute step (4); γ is the angle between the straight line where the pedestrian and the vehicle are located and the direction of vehicle travel; (4) Establish a risk assessment model to determine the probability of a "human-vehicle collision". If the probability is greater than or equal to a threshold value P, execute step (5); otherwise, it is considered that a "human-vehicle collision" will not occur. (5) Two drones are used for early warning, one is deployed above the sidewalk to alert pedestrians, and the other is deployed on the side of the road to alert drivers.
2. The zebra crossing vehicle collision warning method according to claim 1, characterized in that: In step (1), image recognition is performed using the YOLOv5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street. The YOLOv5 algorithm is used to detect vehicles and generate a bounding box for each vehicle. The coordinates of the bounding boxes are used to analyze the positions of these bounding boxes and the degree of overlap, and the spatial relationship between the vehicles is determined. If the bounding box of one vehicle is above or in front of the bounding box of another vehicle, and the bounding boxes of the two vehicles overlap, the field of view of the rear vehicle is blocked by the front vehicle.
3. The zebra crossing vehicle collision warning method according to claim 1, characterized in that: In step (2), the distance L1 between the rear vehicle and the camera and the distance L2 between the pedestrian and the camera are calculated as follows: Where i = 1, 2; C is a constant; BP is the size of the actual object in the picture in the Bpix camera; Connect the camera, pedestrians and vehicles to form a triangle, and calculate the distance S between the pedestrian and the vehicle based on the triangle relationship: Among them, ω is the angle between the vehicle and pedestrian and the camera.
4. The zebra crossing vehicle collision warning method according to claim 3, characterized in that: Among them, R is the radius of the vehicle or pedestrian to be tested; AP is the Apix fixed pixel, that is, the pixel size; α is the field of view angle; The pixel size of the reference object in the image is measured using the YOLOv5 algorithm, and a scaling factor is obtained based on the pixel size and the known actual size of the reference object. This scaling factor is used to convert the pixel size of the vehicle or pedestrian in the image into the actual size. First, determine the bounding box of the reference object in the image and measure the pixel value of its width; let the width of the reference object in the image be P wref pixels, the actual width of the reference object is known, set to W ref ; Use the pixel size of the reference object and the actual size to calculate the width scale factor S w : For the vehicle or pedestrian to be measured in the image, the pixel width P of its bounding box is also measured wobj , the pixel size of the vehicle or pedestrian is converted to the actual size through the above-calculated scaling factor; the actual width of the vehicle or pedestrian W obj Calculated by the following formula: W obj =P wobj ×S w The vehicle or pedestrian radius R passes through the actual width W obj Divide by 2 and you get:
5. The zebra crossing vehicle collision warning method according to claim 1, characterized in that: In step (3), the minimum safe distance model between people and vehicles does not consider the lateral offset, that is, the offset perpendicular to the road direction. The longitudinal safety distance is the minimum safe distance between people and vehicles along the road direction, that is, the longitudinal distance D is calculated. x ; Among them, v v represents the initial speed of the vehicle before braking, t s represents the driver's reaction time, a max Indicates the maximum braking deceleration during braking, v p Indicates the speed of pedestrians when the vehicle brakes, t p It indicates the time for pedestrians to walk after the vehicle decelerates, through v p t p cosθ can be used to calculate the distance traveled longitudinally by the pedestrian, where θ is the heading angle of the vehicle; d s It indicates the safe parking distance, that is, the acceptable safe parking distance between cars and pedestrians.
6. The zebra crossing vehicle collision warning method according to claim 1, characterized in that: In step (4), the environmental impact parameters are used to establish a risk assessment model to determine the potential risk P(γ sk =), environmental impact parameters include drivable channel c r , flow rate under visual obstruction ω0, number of lanes crossed n1, speed of obstruction v0, isolation zone d i , headway time t h and spatiotemporal contexts t ; Driving channel c r The value is assigned by judging whether the lane with the "vehicle-person collision" phenomenon is a lane that traffic participants can drive in accordance with traffic regulations: The flow rate ω0 under obstructed vision is obtained by assigning the flow level of traffic participants around the obstructed area: The number of lanes crossed n1 is assigned by the number of one-way lanes that pedestrians cross after leaving the field of vision obstruction area: The occluder speed v0 is assigned by the speed level of the occluder moving towards the occluded area, where the occluder speed v0 of the static occluder is set to 0: Isolation zone i The value is assigned based on whether there are isolation barriers on the passageway that the traffic participants cross: Headway time t h The time distance between the vehicle in the "vehicle-person collision" lane and the vehicle in front of it in the same lane is assigned: Spatiotemporal contexts t By considering rush hour or places where pedestrians are likely to appear, such as schools and residential areas, the value is assigned as: Potential risks sk By comprehensively considering the seven environmental impact factors, they are classified into two situations: safety and risk, namely: Assume that the average number of people entering the visual field obstruction area per minute is ω. Since the behavior of pedestrians entering the visual field obstruction area is random, the number of pedestrians entering the obstruction area obeys the Poisson distribution. Then the probability of a "pedestrian-vehicle collision" when pedestrians randomly cross the road is: Among them, P(γ sk =1) indicates the probability of a "pedestrian-vehicle collision" when a pedestrian crosses the road; 1-e -ω / 60 represents the probability of pedestrians reaching the visual field obstruction area per unit time; e -ω / 60 =P(N(t=1s))=0 is the probability that no one reaches the blocked area; K ρ is the environmental impact coefficient.
7. The zebra crossing vehicle collision warning method according to claim 6, characterized in that: Among them, k c represents the "probability coefficient of traffic participants crossing the road illegally", which obeys the normal distribution; k d represents the "isolation zone influence coefficient"; k t represents the "headway influence coefficient"; k v Indicates "Obstacle Speed Influence Coefficient".
8. The zebra crossing vehicle collision warning method according to claim 7, characterized in that: k c ~N(0.4,0.02),k d =0.46,k v =1.45,k t =0.05。 9. The zebra crossing vehicle collision warning method according to any one of claims 6 to 8, characterized in that: Configure the threshold P = 0.35, compare P(γ sk =1) and the threshold value P for early warning:
10. A zebra crossing vehicle collision warning system, characterized in that: include: Image acquisition module, used to collect video data of vehicles and pedestrians in the set monitoring area; The image recognition module is used to perform image recognition on the collected vehicle and pedestrian video data using the YOLOv5 algorithm. When a pedestrian appears on the opposite side of the sidewalk, it is determined that there is a pedestrian crossing the street. At the same time, it identifies whether there is a vehicle and determines whether the rear vehicle's field of view is blocked. If there is a situation where the field of vision is blocked, the distance L1 between the vehicle and the camera and the distance L2 between the pedestrian and the camera are calculated by the monocular ranging algorithm, and the distance S between the pedestrian and the vehicle is calculated to determine whether the longitudinal component Scosγ of the distance S between the pedestrian and the vehicle is less than or equal to the longitudinal safety distance D. x ; Risk assessment module, used when the longitudinal component Scosγ of the distance between the person and the vehicle is less than or equal to the longitudinal safety distance D x When the vehicle is in a collision with a human, the probability of the collision is determined according to the risk assessment model. And, the early warning module is used to dispatch two drones for early warning when the probability of a "human-vehicle collision" is greater than or equal to a threshold P. One drone is deployed above the sidewalk to remind pedestrians, and the other is deployed on the side of the road to remind the driver.
Citation Information
Patent Citations
Roadside blind area early warning method and device based on machine vision
CN115862380A
Pedestrian-vehicle conflict analysis method for unsigned pedestrian crossing under mixed traffic flow
CN116187067A
Monocular camera distance measurement method and device, storage medium and electronic equipment
CN116295237A
No-signal zebra crossing man-vehicle interaction risk warning method and system
CN118230597A
Driver-vehicle-road microscopic traffic system modeling and risk identification method and device
WO2023045265A1
