Intelligent vehicle path planning method and system based on AI identification and difference ratio sum algorithm

Abstract

The invention discloses an intelligent vehicle path planning method and system based on AI identification and a difference ratio sum algorithm, and the method comprises the steps: carrying out the data communication of an upper computer and a drive control panel, starting a thread, and collecting an original image; carrying out preprocessing, and carrying out line-by-line scanning on the preprocessed image through a difference ratio and an algorithm to obtain key features of a lane; the method comprises the following steps: carrying out AI identification on an acquired original image through a yov3mobilenet lightweight model to obtain a mark identification result of the image; scene recognition and path planning are carried out on key features of a lane and a sign recognition result of an image through the hierarchical state machine, and path planning information is sent to the lower computer to control the vehicle, so that intelligent vehicle path planning is realized. According to the method, the path planning accuracy can be improved, multi-scene intelligent identification and switching are met, and automatic driving under complex road conditions is supported. The intelligent vehicle path planning method and system based on the AI identification and difference ratio sum algorithm can be widely applied to the technical field of intelligent vehicle path planning.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent vehicle path planning, in particular to an intelligent vehicle path planning method and system based on AI recognition and difference quotient algorithm. BACKGROUND

[0002] An intelligent driving vehicle reconstructs the environment around the vehicle through a vehicle-mounted sensor, obtains the current position and environmental information, and automatically plans a path that can reach the target point. Intelligent driving is a technology integrating artificial intelligence, computer vision, control engineering, algorithm and other disciplines, and represents the development of scientific research strength and industrial level, and has a wide range of applications and development prospects in China. Path planning technology is an important research direction in intelligent driving research, and the main purpose is to enable intelligent vehicles to quickly and efficiently pass through obstacles or travel along a prescribed route.

[0003] At present, a vehicle obtains an RGB image of a front road through a camera, converts the RGB image into a grayscale image through calculation, processes the grayscale image using Otsu method and edge detection algorithm, separates the lane and background in the image to obtain a binary image, and obtains lane boundary information. The driving route of the vehicle is calculated according to the positions of the left and right boundaries of the lane in the binary image.

[0004] However, in the image preprocessing process using Otsu method, if the original image has noise, uneven illumination, a large difference in size between the target and the background, and the like, the grayscale histogram will have no obvious double peaks or multiple peaks, the preprocessing effect is not good, effective lane boundary information cannot be obtained, and effective path planning cannot be performed. In the image preprocessing process using the edge detection algorithm, if the lane boundary part of the original image has noise, the preprocessing effect is not good, effective lane boundary information cannot be obtained, and effective path planning cannot be performed. At the same time, the two binaryzation algorithms need to count the grayscale information of the whole image, and the operation time is long, which is gradually not suitable for the automatic driving system which has higher and higher real-time requirements. If there are obstacles and pedestrians on the lane, the obstacle and pedestrian information is lost in the black and white image after binaryzation, which is not conducive to effective path planning. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide an intelligent vehicle path planning method and system based on AI recognition and difference quotient algorithm, which can improve the accuracy of path planning, meet the intelligent recognition and switching of multiple scenes, and support automatic driving in complex road conditions.

[0006] The first technical solution adopted by the present application is an intelligent vehicle path planning method based on AI recognition and difference quotient algorithm, comprising the following steps: Data communication is performed on an upper computer and a drive control board, three-level thread priority is created and a thread is started, and an original image is collected. The collected original image is preprocessed, and the preprocessed image is scanned line by line through a difference ratio and algorithm to obtain key features of the lane; The original image collected is recognized through a yolov3_mobilenet lightweight model to obtain a sign recognition result of the image; The key features of the lane and the sign recognition result of the image are recognized through a hierarchical state machine, and path planning information is sent to a lower computer to control the vehicle, so as to realize intelligent vehicle path planning.

[0007] Further, the step of communicating data between the upper computer and the drive control board, creating three levels of thread priorities and starting threads, and collecting original images specifically includes: Hardware initialization is performed, a device tree and a kernel module are loaded, a handshake protocol with the drive control board is established through CRC checking, parameters are configured, and data communication between the upper computer and the drive control board is realized; Three levels of thread priorities are created, including real-time threads, high-priority threads and normal threads; Based on the created three levels of thread priorities, zero-copy communication between threads is realized through a ring buffer, AI inference threads and camera acquisition threads are started, and original images are collected.

[0008] Further, the step of preprocessing the collected original image and scanning the preprocessed image line by line through a difference ratio and algorithm to obtain key features of the lane specifically includes: Based on the original image collected by the camera acquisition thread, image correction, image graying and size compression processing are performed through bilinear interpolation and GPU parallel computing to obtain a preprocessed image; The preprocessed image is scanned line by line through a difference ratio and algorithm to identify the lane and obtain key features of the lane.

[0009] Further, the step of scanning the preprocessed image line by line through a difference ratio and algorithm to identify the lane and obtain key features of the lane specifically includes: The preprocessed image is initialized to obtain lane left and right edge point sets, driving path point sets, initial calculation of row midpoint coordinates and forward-looking rows; It is determined whether the point in the lowest row center of the preprocessed image is a point with a gray value close to black, if so, a point with a gray value close to white is found, and the point is taken as the starting point of the first left and right traversal; if not, the point in the lowest row center of the image is taken as the starting point of the first left and right traversal; The point in the center of the lowest row is traversed to the left and right sides of the row of the image with a fixed step size to find two pixel points with a gray value difference greater than a preset threshold, and a difference ratio and operator is defined to perform operation to obtain the left and right boundary information of the lane; The column coordinates of the left and right boundary information points of the lane are added and divided by 2 to obtain the column coordinates of the starting point of the second left and right traversal; All left and right boundary information points of the lane are traversed to construct the left and right boundary point set of the image lane, and the lane width, curvature and continuity features are calculated to obtain the key features of the lane.

[0010] Further, the expression of the difference ratio and operator is specifically as follows:

[0011] In the above formula, represents the gray value of a pixel in a certain row and column of the image, represents a step size, represents another pixel gray value after step size operation in the same row, represents the difference ratio and operator.

[0012] Further, the logo recognition result of the image specifically includes the ID, label, confidence, row coordinate in the image, column coordinate in the image, width and height of the target.

[0013] Further, the scene recognition and path planning are performed on the key features of the lane and the logo recognition result of the image by the hierarchical state machine, and the path planning information is sent to the lower computer to control the vehicle, so as to realize the intelligent vehicle path planning step, which specifically includes: The scene recognition is performed on the key features of the lane and the logo recognition result of the image by the top layer of the hierarchical state machine to obtain a scene recognition result; The behavior decision is performed on the vehicle by the middle layer of the hierarchical state machine to obtain a behavior decision result; Based on the bottom layer of the hierarchical state machine, the scene recognition result and the behavior decision result are combined to generate a driving trajectory to obtain path planning information; The path planning information is sent to the lower computer to control the vehicle to realize intelligent vehicle path planning.

[0014] The second technical solution adopted by the application is: an intelligent vehicle path planning system based on AI recognition and a difference ratio and algorithm, which includes: A first module is used for data communication between the upper computer and the drive control board, creating a three-level thread priority and starting a thread, and collecting an original image; A second module is used for pre-processing the collected original image, and performing row-by-row scanning on the pre-processed image through a difference ratio and algorithm to obtain key features of the lane; The third module is configured to perform AI recognition on the collected original image through the yolov3_mobilenet lightweight model to obtain a sign recognition result of the image. The fourth module is configured to perform scene recognition and path planning on the key features of the lane and the sign recognition result of the image through a hierarchical state machine, and send the path planning information to the lower computer to control the vehicle, thereby realizing intelligent vehicle path planning.

[0015] The method and system have the following advantages: the upper computer and the drive control board are connected for data communication, three-level thread priorities are created and threads are started, original images are collected, the collected original images are preprocessed, the preprocessed images are scanned line by line through the difference and algorithm, the key features of the lane are obtained, the preprocessed images are scanned line by line through the difference and algorithm, the left and right edge point sets of the lane are identified, the adaptive threshold and connectivity analysis are used to eliminate noise points, the continuity and accuracy of the edge point set are ensured, the yolov3_mobilenet lightweight model is used to perform AI recognition on the collected original image to obtain a sign recognition result of the image, the hierarchical state machine is used to perform scene recognition and path planning on the key features of the lane and the sign recognition result of the image, and the path planning information is sent to the lower computer to control the vehicle, thereby realizing intelligent vehicle path planning, combining AI recognition with the difference and algorithm to improve the accuracy of path planning, multi-scene intelligent recognition and switching, and supporting automatic driving in complex road conditions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a step flow chart of the intelligent vehicle path planning method based on AI recognition and the difference and algorithm of the present application; Figure 2 is a structural block diagram of the intelligent vehicle path planning system based on AI recognition and the difference and algorithm of the present application; Figure 3 is a frame schematic diagram of the intelligent vehicle path planning provided by the embodiment of the present application; Figure 4 is a schematic diagram of the AI recognition result provided by the embodiment of the present application; Figure 5 is a schematic diagram of the crossroad line supplement result provided by the embodiment of the present application; Figure 6 is a schematic diagram of the side parking area line supplement result provided by the embodiment of the present application; Figure 7 is a schematic diagram of the fork road line supplement result provided by the embodiment of the present application; Figure 8 is a schematic diagram of the loop line supplement result provided by the embodiment of the present application; Figure 9is a schematic diagram of the framework of the difference ratio and algorithm calculation process provided by the embodiment of the present application. Figure 10 is a schematic diagram of the lane left and right boundaries and the driving route after path planning provided by the embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application will be further described in detail below in combination with the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of description, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0018] Referring to Figure 1 and Figure 3 , the present application provides an intelligent vehicle path planning method based on AI recognition and difference ratio and algorithm, which comprises the following steps: S100, data communication is performed on the host computer and the drive control board, three-level thread priorities are created and threads are started, and original images are collected; Specifically, hardware initialization is performed, device tree and kernel module are loaded, handshake protocol with the drive control board is established by CRC check, parameters are configured, data communication between the host computer and the drive control board is realized; three-level thread priorities are created, including real-time threads, high-priority threads and ordinary threads; based on the created three-level thread priorities, zero-copy communication between threads is realized by using a ring buffer, AI inference threads and camera acquisition threads are started, and original images are collected.

[0019] In this embodiment, first, after the embedded Linux host computer is powered on, hardware initialization (memory, GPIO, clock) is completed, device tree and kernel module are loaded, available USB serial port devices are automatically detected and connected, handshake protocol with the drive control board is established by CRC check, serial port parameters are configured: 115200bps / 8N1, DMA transmission mode, data communication between the host computer and the drive control board is realized.

[0020] Further, three-level thread priorities are created, real-time threads (serial port communication, 1ms period), high-priority threads (AI inference, 10ms period), and ordinary threads (image acquisition, 33ms period), zero-copy communication between threads is realized by using a ring buffer, and a watchdog thread is deployed to monitor thread deadlock. The serial port receiving and sending threads are started, real-time data interaction with the underlying hardware is realized. The AI inference threads and camera acquisition threads are started, and original images are continuously collected.

[0021] S200, the collected original images are preprocessed, and the preprocessed images are scanned row by row by using the difference ratio and algorithm, and the key features of the lane are obtained; S210, based on the original image collected by the camera collection thread, through bilinear interpolation and GPU parallel calculation, image correction, image graying and size compression processing are carried out to obtain the preprocessed image; In this embodiment, image preprocessing is performed. The collected original image is corrected by bilinear interpolation and GPU parallel calculation to eliminate distortion. Then image graying and size compression are performed.

[0022] S220, by difference ratio and algorithm, the preprocessed image is scanned line by line, and the key features of the lane are obtained.

[0023] Specifically, the preprocessed image is initialized, the left and right edge point set of the lane, the driving path point set, the initial calculation row midpoint coordinate and the forward-looking row are obtained. It is determined whether the point in the lowest row center of the preprocessed image is a point with a gray value close to black. If so, find the adjacent point with a gray value close to white, and take the point as the starting point of the first left and right traversal. If not, take the point in the lowest row center of the image as the starting point of the first left and right traversal. Traverse the point in the lowest row center to the left and right sides of the image with a fixed step size, find two pixel points with a gray value difference greater than a preset threshold, and define a difference ratio and operator to perform operation, to obtain the left and right boundary information of the lane. Add the column coordinates of the left and right boundary information points of the lane and divide by 2 to obtain the column coordinates of the starting point of the second left and right traversal. Traverse all the left and right boundary information points of the lane to construct the left and right edge point set of the image lane, calculate the lane width, curvature and continuity features, and obtain the key features of the lane.

[0024] Run the difference ratio and algorithm to scan the preprocessed image line by line to identify the left and right edge point set of the lane. Adopt adaptive threshold, connectivity analysis and other methods to exclude noise points and ensure the continuity and accuracy of the edge point set. According to the left and right edge point set of the lane, the lane width, curvature and continuity are calculated, and the current lane type (straight, curve, missing line, special area, etc.) is judged. The number of white points in the middle column of the image is counted to assist scene switching and function area identification. Detect whether the lane appears left curve or right curve, and realize it through the change trend and curvature analysis of the edge point set. Judge whether one side edge line is continuously lost to trigger line supplement or special processing logic.

[0025] More specifically, as shown in Figure 9 the running difference ratio and algorithm of the embodiment of the present scheme has the following steps: 1) Initialize all variables after the last round of calculation, including the left and right edge point set of the lane, the driving path point set, the initial calculation row midpoint coordinate, the forward-looking row, etc.

[0026] 2) Process the compressed gray image. First determine whether the point in the center of the lowest row of the image is a point with a gray value close to black. If so, find a point with a gray value close to white adjacent to it, and take the point as the starting point of the first left-right traversal. If not, directly take the point in the center of the lowest row of the image as the starting point of the first left-right traversal. Compared with the method of fixing the traversal point, the method of finding the traversal point effectively avoids the shortcoming of failing to correctly search for the left and right boundaries of the lane due to a curve.

[0027] 3) Take the point as the starting point, and traverse to the left and right sides of the row of the image at a fixed step size to find two pixel points with a large difference in gray value. Define the Sum-And-Ratio (SAR) operator:

[0028] wherein is the gray value of a pixel in a certain row and column, is the step size, is the gray value of another pixel in the same row after step size operation, and the threshold value .

[0029] Take the gray values of the two pixels and into to perform operation, when , it is determined as a lane boundary point. If no point meeting the requirements is found in the row, the left and right boundaries of the image in the row are determined as lane boundary points. Through the step size, noise affecting the threshold value in the image is effectively filtered out, and the accuracy of searching for the left and right boundaries of the lane is improved. The left and right boundaries of the lane after the Sum-And-Ratio algorithm calculation are shown in Figure 10 .

[0030] 4) Add the column coordinates of the left and right boundary points of the lane in the row and divide by 2 to obtain the column coordinate of the starting point of the second left-right traversal. At the same time, determine whether there is a case of loss of the left and right boundaries of the lane in the row.

[0031] 5) Continuously perform the above calculation process from the lowest row to the forward-looking row to find the left and right edge point set of the lane in the image. If the point coordinates in the left and right edge point set of the lane coincide, record the row coordinate of the point as the forward-looking row.

[0032] 6) According to the left and right edge point set of the lane, calculate the lane width, curvature, continuity and other features, and determine the current lane type (straight, curve, lost line, special area, etc.). Count the number of white points in the middle column of the image to assist scene switching and function area identification. Analyze whether the lane appears left or right curve through the trend of the edge point set and the curvature.

[0033] S300, AI recognition is performed on the collected original image through the yolov3_mobilenet lightweight model to obtain a sign recognition result of the image. Specifically, AI recognition is performed. The model root directory is passed in, and yolov3_mobilenet lightweight model initialization and loading are performed. After starting, the background inference thread is responsible for calculation after the main thread obtains a new image every frame. Real-time pictures collected by the camera are recognized for vehicles, pedestrians, and traffic signs to obtain the ID, label, confidence, row coordinate in the image, column coordinate in the image, width, and height of the target, as shown in Figure 4

[0034] S400, scene recognition and path planning are performed on the key features of the lane and the sign recognition result of the image through a hierarchical state machine, and the path planning information is sent to the lower computer for controlling the vehicle to realize intelligent vehicle path planning.

[0035] Specifically, scene recognition is performed on the key features of the lane and the sign recognition result of the image through the top layer of the hierarchical state machine to obtain a scene recognition result; behavior decision is performed on the vehicle through the middle layer of the hierarchical state machine to obtain a behavior decision result; based on the bottom layer of the hierarchical state machine, the scene recognition result and the behavior decision result are combined to generate a driving trajectory to obtain path planning information; the path planning information is sent to the lower computer for controlling the vehicle to realize intelligent vehicle path planning.

[0036] In this embodiment, the hierarchical state machine is run in the main thread. The top layer: scene recognition (parking area, straight road acceleration area, slope, loop), the middle layer: behavior decision (following / overlapping / letting), and the bottom layer: driving trajectory generation.

[0037] Road condition recognition is performed based on the AI recognition result and the image preprocessing result, and corresponding path planning is performed according to different road conditions. When the AI recognition result is a crossroad, the state machine switches to a crossroad mode, performs corresponding line filling operations on the image, and re-plans the route to enable the vehicle to pass through smoothly, as shown in Figure 5 When the AI recognition result is a side parking area, the state machine switches to a side parking area, identifies the parking area boundary, performs corresponding line filling operations on the image, and reduces the vehicle speed to enable the vehicle to enter the parking area smoothly, as shown in Figure 6 When the AI recognition result is a fork, the state machine switches to a fork, performs corresponding line filling operations on the image, and guides the vehicle to the correct direction, as shown in Figure 7 When the AI recognition result is a loop, the state machine switches to a loop, performs corresponding line filling operations on the image, and guides the vehicle to the loop, as shown in Figure 8 The serial port sends the path planning information to the lower computer to perform corresponding control on the vehicle controller (rudder, motor) to realize path planning of the vehicle.

[0038] ​Therefore, the embodiment of the present application has the following distinguishing technical features compared with the prior art: 1) AI recognition combined with difference ratio and algorithm greatly improves path planning accuracy, multi-scene intelligent recognition and switching, and supports automatic driving in complex road conditions.

[0039] 2) Compared with the traditional route relying on grayscale-binarization-edge detection, the difference ratio algorithm is used to identify the left and right lane lines through a grayscale image, and the processing speed is faster.

[0040] 3) Developed based on C++ multi-threading, the calculation speed is faster, and the real-time performance of path planning is improved.

[0041] 4) The system has good robustness and is suitable for path planning in most scenarios and can adapt to complex lighting environments.

[0042] 5) The advantages of software and hardware cooperation are fully utilized, and multi-dimensional constraint conditions are met through precise task allocation and scheduling.

[0043] In summary, the embodiment of the present application has the following advantages compared with the prior art: Compared with the traditional route relying on grayscale-binarization-edge detection, the present application directly embeds YOLO AI recognition results into the path planning process: first, the camera gives real-time color bounding boxes and class confidence of obstacles, pedestrians, lane lines and other targets, and then uses the "difference ratio and algorithm" to identify the left and right boundaries of the lane based on the preprocessed grayscale image. Without traversing each pixel point of the image, the joint operation of obstacle avoidance and lane keeping is completed within the same frame of data. Since the original color information is preserved throughout and C++ multi-threading is used for parallel acceleration, the calculation amount is reduced, which not only avoids the loss of key information caused by binarization, but also compresses the single-frame calculation time to the 10 ms level. With the same embedded computing power, the planning success rate is improved from 85% to 98%, significantly improving the real-time performance and robustness of autonomous driving in complex lighting and sudden obstacle scenarios.

[0044] Reference Figure 2 , the intelligent vehicle path planning system based on AI recognition and difference ratio and algorithm, comprising: The first module 201 is used for data communication between the host computer and the drive control board, creating a three-level thread priority and starting the thread, and collecting the original image; The second module 202 is used for pre-processing the collected original image, and performing row-by-row scanning on the pre-processed image through the difference ratio and algorithm to obtain the key features of the lane; The third module 203 is used for AI recognition of the collected original image through the yolov3_mobilenet lightweight model to obtain the image recognition result; The fourth module 204 is configured to perform scene recognition and path planning on the key features of the lane and the sign recognition result of the image by using a hierarchical state machine, and send the path planning information to the lower computer to control the vehicle, so as to realize intelligent vehicle path planning.

[0045] The content in the method embodiments is applicable to the system embodiments, the system embodiments specifically realize the same functions as the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0046] The above is a specific description of the preferred implementation of the application, but the application is not limited to the described embodiments. Those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the application. These equivalent modifications or replacements are all included in the scope defined by the claims of the present application.

Claims

1. An intelligent vehicle path planning method based on AI recognition and difference ratio and algorithm, characterized in that, The method comprises the following steps: The host computer and the drive control board are communicated, three-level thread priorities are created, and threads are started, and original images are collected; The collected original images are preprocessed, and the preprocessed images are scanned line by line through a difference ratio and sum algorithm to obtain key features of the lane; An yolov3_mobilenet lightweight model is used to perform AI recognition on the collected original images to obtain a sign recognition result of the images; A layered state machine is used to perform scene recognition and path planning on the key features of the lane and the sign recognition result of the images, and path planning information is sent to the lower computer to control the vehicle, thereby realizing intelligent vehicle path planning. 2.The AI-based intelligent vehicle path planning method according to claim 1, wherein, The step of communicating the host computer and the drive control board, creating three-level thread priorities and starting threads, and collecting original images specifically comprises: Hardware is initialized, a device tree and a kernel module are loaded, a handshake protocol with the drive control board is established through CRC checking, parameters are configured, and data communication between the host computer and the drive control board is realized; Three-level thread priorities are created, including real-time threads, high-priority threads and ordinary threads; Based on the created three-level thread priorities, a ring buffer is used to realize zero-copy communication between threads, AI inference threads and camera collection threads are started, and original images are collected. 3.The AI-based intelligent vehicle path planning method of claim 2, wherein, The step of preprocessing the collected original images and scanning the preprocessed images line by line through a difference ratio and sum algorithm to obtain key features of the lane specifically comprises: Based on the original images collected by the camera collection thread, image correction, image graying and size compression processing are performed through bilinear interpolation and GPU parallel computing to obtain preprocessed images; The preprocessed images are scanned line by line through a difference ratio and sum algorithm to identify the lane and obtain key features of the lane.

4. The AI-based intelligent vehicle path planning method of claim 3, wherein, The step of scanning the preprocessed images line by line through a difference ratio and sum algorithm to identify the lane and obtain key features of the lane specifically comprises: The preprocessed images are initialized to obtain lane left and right edge point sets, driving path point sets, initial calculation row midpoint coordinates and a forward-looking row; It is determined whether the point in the lowest row center of the preprocessed image is a point with a gray value close to black, if so, a point with a gray value close to white adjacent to the point is found as the starting point of the first left and right traversal, if not, the point in the lowest row center of the image is taken as the starting point of the first left and right traversal; The point in the lowest row center is traversed to the left and right sides of the image at a fixed step to find two pixel points with a gray value difference greater than a preset threshold, and a difference ratio and sum operator is defined to perform operation to obtain lane left and right boundary information; The column coordinates of the lane left and right boundary information points are added and divided by 2 to obtain the column coordinates of the starting point of the second left and right traversal; All lane left and right boundary information points are traversed to construct image lane left and right edge point sets, calculate lane width, curvature and continuity features, and obtain key features of the lane. 5.The AI-based intelligent vehicle path planning method using difference ratio and algorithm identification according to claim 4, wherein, The expression of the difference ratio and sum operator is specifically as follows: ； In the above formula, represents the gray value of a pixel in a certain row and column of the image, represents the step size, represents the gray value of another pixel in the same row after the step size operation, represents the difference ratio and operator. 6.The AI-based intelligent vehicle path planning method of claim 5, wherein, The logo recognition result of the image specifically includes an ID, a label, a confidence, a row coordinate in the image, a column coordinate in the image, a width, and a height of the target.

7. The AI-based intelligent vehicle path planning method of claim 6, wherein, The step of performing scene recognition and path planning on the key features of the lane and the logo recognition result of the image by the hierarchical state machine and sending path planning information to a lower computer to control the vehicle to realize intelligent vehicle path planning specifically includes: performing scene recognition on the key features of the lane and the logo recognition result of the image by a top layer of the hierarchical state machine to obtain a scene recognition result; performing behavior decision on the vehicle by a middle layer of the hierarchical state machine to obtain a behavior decision result; generating a driving track based on a bottom layer of the hierarchical state machine in combination with the scene recognition result and the behavior decision result to obtain path planning information; sending the path planning information to the lower computer to control the vehicle to realize intelligent vehicle path planning.

8. An intelligent vehicle path planning system based on AI recognition and difference ratio and algorithm, characterized in that, The method comprises the following modules: a first module for performing data communication on the upper computer and the drive control board, creating three levels of thread priorities and starting threads, and collecting original images; a second module for performing preprocessing on the collected original images, and performing line-by-line scanning on the preprocessed images by a difference ratio algorithm to obtain the key features of the lane; a third module for performing AI recognition on the collected original images by a yolov3_mobilenet lightweight model to obtain the logo recognition result of the image; a fourth module for performing scene recognition and path planning on the key features of the lane and the logo recognition result of the image by the hierarchical state machine, and sending path planning information to the lower computer to control the vehicle to realize intelligent vehicle path planning.

Images