A mobile roadside pedestrian traffic conflict monitoring and early warning device and method

Abstract

The present invention proposes a mobile roadside pedestrian traffic conflict monitoring and early warning device and method, which belongs to the field of road traffic safety technology; it comprises a chassis; a visual recognition unit, which is arranged below the chassis and faces the direction of a blind spot, and is used to obtain an image of the blind spot, identify and obtain motion trajectory information of a target in the image; a main control unit, which is arranged in the chassis and is communicatively connected with the visual recognition module, and is used to receive the motion trajectory information of the target in the image obtained by the visual recognition module, and generate a warning instruction when it is judged that the target enters a preset conflict area; an RGB warning unit, which is arranged on one side of the chassis, has one end hinged with the chassis, and the other end extends horizontally toward the top of the road on the side facing the motor vehicle, and the RGB warning unit is communicatively connected with the main control unit, and is used to receive the warning instruction and display it to the vehicle driver; a system power supply unit, which is arranged above the chassis and is used to provide a working power supply.

Description

Technical Field

[0001] The present invention relates to the technical field of road traffic safety, and in particular to a mobile roadside pedestrian traffic conflict monitoring and early warning device and method. Background Art

[0002] It is very necessary to provide a mobile roadside pedestrian traffic conflict monitoring and warning device that can realize real-time blind spot pedestrian conflict monitoring and warning at day and night in blind spots such as road corner obstructed sections and road reconstruction and expansion fence areas. It is used to improve the blind spots of intersections, provide observable signs, and reduce the possibility of accidents. Summary of the Invention

[0003] In view of this, the present invention proposes a mobile roadside pedestrian traffic conflict monitoring and warning device and method that can effectively monitor pedestrians or non-motor vehicle drivers in blind spots and promptly remind drivers of potential traffic conflict risks.

[0004] The technical solution of the present invention is achieved as follows:

[0005] In one aspect, the present invention provides a mobile roadside pedestrian traffic conflict monitoring and warning device, comprising:

[0006] The chassis is set at a corner where there is a blind spot;

[0007] The visual recognition unit is arranged under the chassis and faces the blind area of ​​the field of view, and is used to obtain images of the blind area of ​​the field of view and identify and obtain the motion trajectory information of the target in the image;

[0008] The main control unit is provided in the chassis and is in communication with the visual recognition module, and is used to receive the motion trajectory information of the target in the image obtained by the visual recognition module, and generate a warning instruction when it determines that the target enters a preset conflict area;

[0009] An RGB warning unit is provided on one side of the chassis, with one end hinged to the chassis and the other end extending horizontally above the road facing the side where the motor vehicle is traveling. The RGB warning unit is in communication with the main control unit and is used to receive warning instructions and display them to the vehicle driver;

[0010] The system power supply unit is arranged above the chassis and is electrically connected to the visual recognition module, main control unit and RGB warning unit respectively to provide working power.

[0011] On the basis of the above scheme, preferably, the visual recognition unit includes an infrared night vision camera and a visual processing chip, and the infrared night vision camera is communicatively connected to the visual processing chip; the infrared night vision camera obtains an image of the blind spot and sends the image to the visual processing chip; the visual processing chip identifies the target position in the image, and combines the images at different times to mark the target's motion trajectory, obtains the target's real-world coordinates through perspective conversion, and determines whether the target enters a preset conflict area. If so, the motion trajectory information including the target's position, distance and speed is sent to the main control unit; otherwise, the visual recognition unit will not communicate with the main control unit, and the visual processing chip will re-analyze the acquired image of the blind spot.

[0012] Preferably, the main control unit determines whether the target enters the preset conflict area through the determination conditions including the willingness degree I and the target distance D. When both the willingness degree I and the target distance D meet the set requirements at the same time, it is considered that the target is about to enter the preset conflict area; otherwise, the visual processing chip re-analyzes the image of the blind spot; wherein the target is a pedestrian and a non-motor vehicle driver.

[0013] More preferably, the willingness I is calculated by the following formula:

[0014] Where l is the general trajectory line of the road corner; t represents the current moment. At the current moment t, ten frames are extracted from the image of the blind spot that has been collected, and two different frames are extracted every second. (x t-i ,y t-i ) to represent the real-world coordinates of the target in different frames, i = 0, 1, 2, ..., 9, (x t-i ,y t-i )∈{(x t ,y t ), (x t-1 ,y t-1 ),...,(x t-9 ,y t-9 )}; The distance between the general trajectory line l and the target's motion trajectory composed of the real-world coordinates corresponding to different frames; Indicates the degree of deviation between the target's motion trajectory composed of the real-world coordinates corresponding to different frames and the general trajectory line l; the denominator is the weight term, j = 0, 1, 2, ..., 9. The closer to the current time t, the larger the weight term value; v i Represents the target's speed between two adjacent frames. The initial speed is 0.5-1m / s. The larger the value, the greater the possibility of passing through the blind spot. <l,(x t-i -x t-i-1 ,y t-i -yt-i-1 )> represents the cosine similarity between the motion trajectory of the target formed by the real-world coordinates corresponding to any two adjacent frames among the extracted ten frames and the general trajectory line l. The cosine similarity is used to measure the angular difference between the motion trajectory of the target formed by the real-world coordinates corresponding to the target in different frames and the general trajectory line l.

[0015] Further preferably, both the willingness degree I and the target distance D satisfy the set requirements. The specific content is as follows: 1) When the vision processing chip determines that the target distance D < 15m, the vision processing chip starts to collect data immediately. The vision processing chip sends the motion trajectory information of the target to the main control unit, and the main control unit calculates the willingness degree I, but the main control unit will not generate a warning instruction; 2) When the vision processing chip determines that 5m < D < 10m, the vision processing chip sends the motion trajectory information of the target to the main control unit. When the willingness degree I calculated by the main control unit is not less than the set threshold α, the main control unit generates a warning instruction and sends it to the RGB warning unit, and updates the distance between the target and the preset conflict area at regular intervals; when the willingness degree I calculated by the main control unit is less than the set threshold α, the main control unit does not generate a warning instruction; 3) When the vision processing chip determines that the target distance D < 5m, the vision processing chip sends the motion trajectory information of the target to the main control unit. When the willingness degree I calculated by the main control unit > 0, the main control unit generates a warning instruction and sends it to the RGB warning unit.

[0016] Even more preferably, the RGB warning unit includes a rectangular light board; after receiving the warning instruction from the main control unit, the RGB warning unit generates a pattern of a pedestrian in the light board area. The horizontal length of the light board area is proportional to the target distance D, and the distance between the pattern of the pedestrian and the edge of the light board area corresponds to the real-time value of the target distance D; the RGB warning unit changes the relative position of the pattern of the pedestrian on the light board according to the distance between the target and the preset conflict area updated by the main control unit at regular intervals.

[0017] Even further preferably, the main body light-emitting parts are arranged in an array on the rectangular light board, and the main body light-emitting parts use yellow LED lights.

[0018] Preferably, the target distance D is obtained by calculating the target distance D through the visual processing chip after the infrared night vision camera acquires the 940nm infrared band image, and the target distance D is written into the target's motion trajectory information; the visual processing chip is integrated with the YOLOV5s algorithm based on lightweight modification, and the collected infrared band images are manually annotated to form a data set, and the data set is divided into a training set and a validation set, and the training set is input into the lightweight modified YOLOv5s algorithm, and a model for identifying targets in infrared band images and measuring the target distance D is obtained through training; the YOLOV5s algorithm based on lightweight modification is formed by improving the Conv module of the Backbone part of the original YOLOV5 algorithm by introducing the GhostNet module, and then introducing the FasterNet module to replace the structure of the Backbone part of the original YOLOV5 algorithm.

[0019] On the basis of the above technical solution, preferably, the system power supply unit includes a rechargeable battery, a photovoltaic power generation component, an AC charging interface and a low-voltage DC charging interface, and the photovoltaic power generation component, the AC charging interface and the low-voltage DC charging interface are all electrically connected to the rechargeable battery; the rechargeable battery is electrically connected to the visual recognition module, the main control unit and the RGB warning unit respectively.

[0020] On the other hand, the present application also provides a mobile roadside pedestrian traffic conflict monitoring and early warning method, comprising the following steps:

[0021] S1: equipped with the above-mentioned mobile roadside pedestrian traffic conflict monitoring and warning device;

[0022] S2: The visual recognition module is trained using the lightweight modified YOLOV5s algorithm to obtain a model for identifying targets and measuring target distance D.

[0023] S3: After acquiring the image of the blind spot, the visual recognition module inputs it into the model for identifying the target and calculating the target distance D. It takes the pedestrian or non-motor vehicle driver in the image as the target, obtains the target's motion trajectory information, and sends it to the main control unit;

[0024] S4: After receiving the target's motion trajectory information, the main control unit determines whether the target has entered the preset conflict area. If so, the main control unit generates a warning instruction and sends it to the RGB warning unit for output display. If not, the visual recognition module re-analyzes the image of the blind spot.

[0025] S5: After receiving the warning command, the RGB warning unit generates a pedestrian pattern based on the distance between the target and the preset conflict area, and displays it to the vehicle driver through a light display, reminding the driver that there is a target approaching in the blind spot and to slow down and avoid it in time;

[0026] S6: When the target leaves the monitoring area of ​​the visual recognition module or passes through the set conflict area, the display content of the RGB warning unit is reset and steps S3-S5 are repeated.

[0027] The present invention provides a mobile roadside pedestrian traffic conflict monitoring and early warning device and method, which has the following advantages over the prior art:

[0028] 1) This application addresses the problem of blind spots that are undetectable to vehicle drivers. The mobile roadside pedestrian traffic conflict monitoring and warning device acquires images of the blind spots, identifies pedestrians or non-motor vehicle drivers in the images as targets, and obtains the target's motion trajectory information through analysis. When the target approaches a preset conflict area, the main control unit generates a warning instruction and activates the RGB warning unit to provide a warning light prompt, allowing drivers to identify blind spot risks and reduce the risk of traffic accidents.

[0029] 2) This application utilizes an infrared night vision camera in combination with a visual processing chip to achieve visual recognition. The optimization of the YOLOv5 Backbone component through the GhostNet and FasterNet modules significantly improves the image processing speed, thereby increasing the product's computer vision processing speed from 8.5fps to 15fps, improving the real-time performance and effectiveness of the detection and warning device.

[0030] 3) This application proposes a method for calculating the willingness to pass, which fully combines the degree of deviation between the target's motion trajectory and the general trajectory line formed by the real-world coordinates corresponding to different frames, the target's speed between two adjacent frames, the cosine similarity and weight terms between the target's motion trajectory and the general trajectory line, and fully considers the target's position, movement speed and movement direction. It can effectively measure the target's passing tendency and well judge whether the target has passed. It can also predict whether the target has passed through the preset conflict area near the blind spot; if the target has a strong willingness to pass, the main control unit generates a warning command and triggers the RGB warning unit; if the target enters the camera image without the purpose of passing through the preset conflict area, the main control unit will not generate a warning command;

[0031] 4) The mobile roadside pedestrian traffic conflict monitoring and warning device is convenient and universal. This device can be detachably mounted on a road guardrail and is ready for operation. Considering the inconvenience of power supply in most road reconstruction and expansion scenarios, and to meet the needs of all-weather use, this product is designed with multiple power supply methods. The power is supplied by a Type-C socket, solar panels, or 220V AC to a rechargeable battery, which then powers the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 This is a structural block diagram of a mobile roadside pedestrian traffic conflict monitoring and early warning device and method of the present invention;

[0034] Figure 2 This is a structural schematic diagram of a chassis of a mobile roadside pedestrian traffic conflict monitoring and early warning device and method of the present invention;

[0035] Figure 3 This is a three-dimensional diagram of the circuit assembly state of a mobile roadside pedestrian traffic conflict monitoring and early warning device and method of the present invention;

[0036] Figure 4 Schematic diagram of calculation of passing willingness in a mobile roadside pedestrian traffic conflict monitoring and early warning device and method according to the present invention under a bird's-eye view and a camera view;

[0037] Figure 5 A flow chart showing the determination state of passing willingness and target distance of a mobile roadside pedestrian traffic conflict monitoring and early warning device and method according to the present invention;

[0038] Figure 6 This is a schematic diagram of the animation design process of the RGB warning unit of a mobile roadside pedestrian traffic conflict monitoring and early warning device and method of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] like Figure 1 、 Figure 2 and Figure 3 As shown, the present invention provides a mobile roadside pedestrian traffic conflict monitoring and warning device, including: a chassis, a visual recognition unit, a main control unit, an RGB warning unit and a system power supply unit. These components are introduced below.

[0041] The chassis is set at a corner position where there is a blind spot in the field of vision; the main body of the chassis can be made by sheet metal welding or can be integrally formed by 3D printing, and a chassis cover for sealing is set on one side of the main body of the chassis.

[0042] The visual recognition unit is arranged below the chassis, facing the direction of the blind spot, and is used to obtain images of the blind spot, identify and obtain the motion trajectory information of the target in the image. Among them, the motion trajectory information of the target includes the line connecting the real-world coordinates obtained by converting the position of the target in the image at different times, the movement speed of the target, and the distance between the target and the preset conflict area in the real-world coordinate system. Because the obtained image is two-dimensional and the real-world coordinate system is three-dimensional, the position of the target in the image coordinate system needs to be converted to the real-world coordinate system through coordinate transformation. This process is the calibration process, which is common knowledge in this field and will not be repeated here. This step can be subdivided into two steps: target recognition and target motion trajectory acquisition. The target here refers to pedestrians and non-motor vehicle drivers in the corner blind spot relative to the side of the vehicle driver.

[0043] In this solution, the visual recognition unit includes an infrared night vision camera and a visual processing chip, which are connected to each other. The infrared night vision camera captures images of blind spots and sends them to the visual processing chip. The visual processing chip identifies the target position in the image and, combining images from different times, marks the target's motion trajectory. By converting the perspective, it obtains the target's real-world coordinates and sends the motion trajectory information, including the target's position, distance, and speed, to the main control unit. During the day, the infrared night vision camera captures daytime images in the 940nm infrared band. At night, the infrared night vision camera, combined with a 940nm infrared fill light, captures nighttime images. The infrared night vision camera can be a dome camera or a box camera. The K210 chip, with a maximum computing power of 1 tops, is used as the visual processing chip. This K210 chip integrates a lightweight modified YOLOV5s algorithm. Collected infrared images are manually annotated to form a dataset, which is then divided into a training set and a validation set. The training set is then fed into the lightweight modified YOLOV5s algorithm, which trains a model for identifying targets and measuring target distances (D) in infrared images. The lightweight modified YOLOV5s algorithm is formed by improving the Conv module in the backbone of the original YOLOV5 algorithm by introducing the GhostNet module, and then replacing the backbone structure of the original YOLOV5 algorithm with the FasterNet module. The lightweight modified YOLOV5s algorithm is a common technique in this field and has open source code, so it will not be detailed here.

[0044] The main control unit, housed within the chassis and connected to the visual recognition module, receives information about the target's motion trajectory from images captured by the module and generates a warning when it determines the target has entered a pre-set conflict zone. In this solution, the main control unit utilizes the PICO main control board. The pre-set conflict zone is a virtual boundary set within the infrared night vision camera's field of view. It is an area adjacent to the corner of the blind spot and adjacent to the roadway on the side of the vehicle. If a target enters this zone, it is highly likely to "ghost," meaning it crosses into the roadway.

[0045] In this solution, the main control unit determines whether the target has entered the preset conflict area by the determination conditions including willingness I and target distance D. When both willingness I and target distance D meet the set requirements, it is considered that the target is about to enter the preset conflict area; otherwise, the visual processing chip re-analyzes the image of the blind spot; the target distance D is obtained by the infrared night vision camera after the infrared band image of 940nm is obtained, and the target distance D is written into the target's motion trajectory information. Figure 4 , calculated by the willingness I using the following formula:

[0046] Where l is the general trajectory line of the road corner; t represents the current moment. At the current moment t, ten frames are extracted from the image of the blind spot that has been collected, and two different frames are extracted every second. (x t-i ,y t-i ) to represent the real-world coordinates of the target in different frames, i = 0, 1, 2, ..., 9, (x t-i ,y t-i )∈(x t ,y t ), (x t-1 ,y t-1 ),...,(x t-9 ,y t-9 )}; The distance between the general trajectory line l and the target's motion trajectory composed of the real-world coordinates corresponding to different frames; Indicates the degree of deviation between the target's motion trajectory composed of the real-world coordinates corresponding to different frames and the general trajectory line l; the denominator is the weight term, j = 0, 1, 2, ..., 9. The closer to the current time t, the larger the weight term value; v i Represents the target's speed between two adjacent frames. The initial speed is 0.5-1m / s. The larger the value, the greater the possibility of passing through the blind spot. <l,(x t-1 -x t-i-1 ,y t-i -y t-i-1)> represents the cosine similarity between the motion trajectory of the target formed by the real-world coordinates corresponding to any two adjacent frames among the extracted ten frames and the general trajectory line l. The cosine similarity is used to measure the angular difference between the motion trajectory of the target formed by the real-world coordinates corresponding to different frames and the general trajectory line l. By the willingness I, it fully combines the deviation degree between the motion trajectory of the target formed by the real-world coordinates corresponding to different frames and the general trajectory line, the speed of the target between two adjacent frames, the cosine similarity between the motion trajectory of the target and the general trajectory line, and the weight term, fully considering the position, moving speed, and moving direction of the target, and can effectively measure the tendency of the target to enter the preset conflict area.

[0047] Reference Figure 5 , when both the willingness I and the target distance D meet the set requirements, it includes the following situations: 1) When the vision processing chip determines that the target distance D < 15m, the vision processing chip starts to collect data, the vision processing chip sends the motion trajectory information of the target to the main control unit, and the main control unit calculates the willingness I, but the main control unit does not generate a warning instruction; 2) When the vision processing chip determines that 5m < D < 10m, the vision processing chip sends the motion trajectory information of the target to the main control unit. When the willingness I calculated by the main control unit is not less than the set threshold α, the main control unit generates a warning instruction and sends it to the RGB warning unit, and updates the distance between the target and the preset conflict area at regular intervals; when the willingness I calculated by the main control unit is less than the set threshold α, the main control unit does not generate a warning instruction; 3) When the vision processing chip determines that the target distance D < 5m, the vision processing chip sends the motion trajectory information of the target to the main control unit. When the willingness I calculated by the main control unit > 0, the main control unit generates a warning instruction and sends it to the RGB warning unit.

[0048] The RGB warning unit is set on one side of the chassis, one end is hinged to the chassis, and the other end horizontally extends upward towards the road on the advancing side of the motor vehicle. The RGB warning unit is communicatively connected to the main control unit and is used to receive warning instructions and display them to the vehicle driver.

[0049] Reference Figure 6 , in this solution, the RGB warning unit includes a rectangular lamp board; after receiving the warning instruction from the main control unit, the RGB warning unit generates a pattern of a pedestrian in the lamp board area. The horizontal length of the lamp board area is proportional to the target distance D, and the distance between the pattern of the pedestrian and the edge of the lamp board area corresponds to the real-time value of the target distance D; the RGB warning unit changes the relative position of the pattern of the pedestrian on the lamp board according to the distance between the target and the preset conflict area updated by the main control unit at regular intervals.

[0050] The rectangular light panel is arrayed with a main light-emitting unit, utilizing yellow LEDs. This is compatible with traffic light color schemes: red indicates a no-go zone, yellow a warning, and green indicates a pass. In addition to the main light-emitting unit, the RGB warning unit can also be configured with a text display area, using red text to remind drivers to slow down and avoid pedestrians and non-motorized vehicles in the corner blind spot. It also displays the target's real-time distance to the pre-set conflict zone.

[0051] For example, Figure 6 As shown, the rectangular light panel is 1 meter long. When a pedestrian or non-motor vehicle in the corner blind spot is 10 meters away from the preset conflict zone boundary, the RGB warning unit generates a yellow pedestrian image according to the warning instruction, and the pedestrian image is displayed at one end of the light panel area. When the pedestrian or non-motor vehicle in the corner blind spot is 5 meters away from the preset conflict zone boundary, the pedestrian image moves to the center of the light panel area. When the pedestrian or non-motor vehicle in the corner blind spot is within 1 meter of the preset conflict zone boundary, the pedestrian image moves to the other end of the light panel area. This display process changes with the actual distance of the pedestrian or non-motor vehicle in the corner blind spot, notifying the driver of the movement direction and relative position of the pedestrian or non-motor vehicle in the corner blind spot, forming a dynamic image display to remind the driver to pay attention to the target's direction and route. The display state of the rectangular light panel of the RGB warning unit resets 1-5 seconds after the pedestrian or non-motor vehicle leaves the field of view of the infrared night vision camera.

[0052] The system power supply unit, located above the chassis, is electrically connected to the visual recognition module, main control unit, and RGB warning unit to provide operating power. The system power supply unit includes a rechargeable battery, photovoltaic panels, an AC charging port, and a low-voltage DC charging port. The photovoltaic panels, AC charging port, and low-voltage DC charging port are all electrically connected to the rechargeable battery; the rechargeable battery is also electrically connected to the visual recognition module, main control unit, and RGB warning unit. Because power supply may be inconvenient for mobile roadside pedestrian conflict monitoring and warning devices installed at roadside barriers during road reconstruction and expansion, the device is equipped with various power supply methods to enhance its adaptability. The AC charging port uses a 220V AC interface, charging the rechargeable battery via a transformer and a step-down chip. The low-voltage DC charging port uses a TYPC-C interface, employing the PD protocol and supporting DC charging with a wide input voltage. The photovoltaic panels generate electricity during daytime using sunlight and store it in the rechargeable battery. The rechargeable battery can be a 3.7V lithium battery or a 12V lead-acid battery.

[0053] Depend on Figure 2 and Figure 3As can be seen, the device is compact and easy to carry, install and dismantle. A hinge is provided on one side of the chassis to facilitate adjustment of the rotation angle of the RGB warning unit, making it easy to flip and store the RGB warning unit, reducing the size of the device during transportation.

[0054] On the other hand, the present application also provides a mobile roadside pedestrian traffic conflict monitoring and early warning method, comprising the following steps:

[0055] S1: equipped with the above-mentioned mobile roadside pedestrian traffic conflict monitoring and warning device;

[0056] S2: The visual recognition module is trained using the lightweight modified YOLOV5s algorithm to obtain a model for identifying targets and measuring target distance D.

[0057] S3: After acquiring the image of the blind spot, the visual recognition module inputs it into the model for identifying the target and calculating the target distance D. It takes the pedestrian or non-motor vehicle driver in the image as the target, obtains the target's motion trajectory information, and sends it to the main control unit;

[0058] S4: After receiving the target's motion trajectory information, the main control unit determines whether the target has entered the preset conflict area. If so, the main control unit generates a warning instruction and sends it to the RGB warning unit for output display. If not, the visual recognition module re-analyzes the image of the blind spot.

[0059] S5: After receiving the warning command, the RGB warning unit generates a pedestrian pattern based on the distance between the target and the preset conflict area, and displays it to the vehicle driver through a light display, reminding the driver that there is a target approaching in the blind spot and to slow down and avoid it in time;

[0060] S6: When the target leaves the monitoring area of ​​the visual recognition module or passes through the set conflict area, the display content of the RGB warning unit is reset and steps S3-S5 are repeated.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A mobile roadside pedestrian traffic conflict monitoring and early warning device, characterized in that: include: The chassis is set at a corner where there is a blind spot; The visual recognition unit is arranged under the chassis and faces the direction of the blind area of ​​the visual field, and is used to obtain images of the blind area of ​​the visual field, and to identify and obtain the motion trajectory information of the target in the image; The main control unit is arranged in the chassis and is in communication connection with the visual recognition module, and is used to receive the motion trajectory information of the target in the image obtained by the visual recognition module, and generate a warning instruction when it is determined that the target enters a preset conflict area; An RGB warning unit is arranged on one side of the chassis, one end of which is hinged to the chassis, and the other end of which is horizontally extended toward the upper side of the road facing the side where the motor vehicle is traveling. The RGB warning unit is connected to the main control unit for communication, and is used to receive warning instructions and display them to the vehicle driver; The system power supply unit is arranged above the chassis and is electrically connected to the visual recognition module, the main control unit and the RGB warning unit respectively to provide working power; The visual recognition unit includes an infrared night vision camera and a visual processing chip, and the infrared night vision camera is communicatively connected with the visual processing chip; the infrared night vision camera obtains an image of the blind area of ​​the visual field and sends the image to the visual processing chip; the visual processing chip identifies the target position in the image, and combines the images at different times to mark the motion trajectory of the target, obtains the real world coordinates of the target through perspective conversion, and determines whether the target enters a preset conflict area. If so, the motion trajectory information including the position, distance and speed of the target is sent to the main control unit; otherwise, the visual recognition unit will not communicate with the main control unit, and the visual processing chip will re-analyze the acquired image of the blind area of ​​the visual field; The main control unit determines whether the target enters the preset conflict area by determining whether the target enters the preset conflict area, including the passing willingness degree I and the target distance D. When both the passing willingness degree I and the target distance D meet the set requirements at the same time, it is considered that the target is about to enter the preset conflict area; otherwise, the visual processing chip re-analyzes the image of the blind spot; wherein the target is a pedestrian and a non-motor vehicle driver; The willingness I is calculated by the following formula: Where l is the general trajectory of the road corner; t represents the current time. At the current time t, ten frames are extracted from the image of the blind spot that has been collected, and two different frames are extracted every second. (x t-i ,y t-i ) to represent the real world coordinates of the target in different frames, i = 0, 1, 2, ..., 9, (x t-i ,y t-i )∈{(x t ,y t ), (x t-1 ,y t-1 ),...,(x t-9 ,y t-9 )}; It expresses the distance between the general trajectory line l and the target's motion trajectory composed of the real-world coordinates corresponding to the target in different frames; Indicates the degree of deviation between the target's motion trajectory composed of the real-world coordinates corresponding to different frames and the general trajectory line l; the denominator is the weight item, j = 0, 1, 2, ..., 9. The closer to the current time t, the larger the weight item value; v i Represents the speed of the target between two adjacent frames. The initial speed is 0.5-1m / s. The larger the value, the greater the possibility of passing through the blind spot. <l,(x t-i -x t-i-1 ,y t-i -y t-i-1 )> represents the cosine similarity between the motion trajectory of the target formed by the real-world coordinates corresponding to any two adjacent frames in the ten extracted frames and the general trajectory line l. The cosine similarity is used to measure the angular difference between the motion trajectory of the target formed by the real-world coordinates corresponding to different frames and the general trajectory line l.

2. A mobile roadside pedestrian traffic conflict monitoring and early warning device according to claim 1, characterized in that: The passing willingness I and the target distance D both meet the set requirements at the same time, and the specific contents are as follows: 1) When the visual processing chip determines that the target distance D is less than 15m, the visual processing chip starts to collect data, and the visual processing chip sends the target's motion trajectory information to the main control unit, and the main control unit calculates the passing willingness I, but the main control unit does not generate a warning instruction; 2) When the visual processing chip determines that the target distance is 5m<D<10m, the visual processing chip sends the target's motion trajectory information to the main control unit, and when the main control unit calculates that the passing willingness I is not less than the set threshold α, the main control unit generates a warning instruction and sends it to the RGB warning unit, and updates the distance between the target and the preset conflict area at regular intervals; When the main control unit calculates that the passing willingness I is less than the set threshold α, the main control unit does not generate a warning instruction; 3) When the visual processing chip determines that the target distance D is less than 5m, the visual processing chip sends the target's motion trajectory information to the main control unit. When the main control unit calculates that the passing willingness I is greater than 0, the main control unit generates a warning instruction and sends it to the RGB warning unit.

3. A mobile roadside pedestrian traffic conflict monitoring and early warning device according to claim 2, characterized in that: The RGB warning unit includes a rectangular light panel; After receiving the warning command from the main control unit, the RGB warning unit generates a pedestrian pattern in the light board area, the horizontal length of the light board area is proportional to the target distance D, and the distance of the pedestrian pattern from the edge of the light board area corresponds to the real-time value of the target distance D; The RGB warning unit changes the relative position of the pedestrian pattern on the light board according to the distance between the target and the preset conflict area updated by the main control unit at regular intervals.

4. A mobile roadside pedestrian traffic conflict monitoring and early warning device according to claim 3, characterized in that: The rectangular light panel is provided with main body light emitting parts in an array, and the main body light emitting parts adopt yellow LED lights.

5. The mobile roadside pedestrian traffic conflict monitoring and early warning device according to claim 1 is characterized in that: The target distance D is obtained by the vision processing chip after the infrared night vision camera obtains the 940nm infrared band image, and the target distance D is written into the target's motion trajectory information; the visual processing chip integrates the YOLOV5s algorithm based on lightweight transformation, and the collected infrared band images are manually annotated to form a data set, and the data set is divided into a training set and a verification set. The training set is input into the lightweight transformed YOLOv5s algorithm, and a model for identifying targets in infrared band images and measuring the target distance D is obtained through training; the YOLOV5s algorithm based on lightweight transformation is formed by introducing the GhostNet module to improve the Conv module of the Backbone part of the original YOLOV5 algorithm, and then introducing the FasterNet module to replace the structure of the Backbone part of the original YOLOV5 algorithm.

6. The mobile roadside pedestrian traffic conflict monitoring and early warning device according to claim 1, characterized in that: The system power supply unit includes a rechargeable battery, a photovoltaic power generation component, an AC charging interface and a low-voltage DC charging interface. The photovoltaic power generation component, the AC charging interface and the low-voltage DC charging interface are all electrically connected to the rechargeable battery; the rechargeable battery is electrically connected to the visual recognition module, the main control unit and the RGB warning unit respectively.

7. A mobile roadside pedestrian traffic conflict monitoring and early warning method, characterized in that: The steps include: S1: A mobile roadside pedestrian traffic conflict monitoring and warning device as described in any one of claims 1 to 6 is configured; S2: The visual recognition module is trained using the YOLOV5s algorithm based on lightweight modification to obtain a model for identifying targets and measuring target distance D; S3: After acquiring the image of the blind spot, the visual recognition module inputs it into the model for identifying the target and measuring the target distance D, takes the pedestrian or non-motor vehicle driver in the image as the target, acquires the target's motion trajectory information and sends it to the main control unit; S4: After receiving the target's motion trajectory information, the main control unit determines whether the target enters the preset conflict area; if so, the main control unit generates a warning instruction and sends it to the RGB warning unit for output display; if not, the visual recognition module re-analyzes the image of the blind spot; S5: After receiving the warning command, the RGB warning unit generates a pedestrian pattern according to the distance between the target and the preset conflict area, and displays it to the vehicle driver through a light display, prompting the driver that there is a target approaching in the blind spot and the driver should slow down and avoid it in time; S6: When the target leaves the monitoring area of ​​the visual recognition module or passes through the set conflict area, the display content of the RGB warning unit is reset and steps S3-S5 are repeated.

Images