Development And Application Of Imaging ADAS System Based On Vehicle SOC

With the rapid development of artificial intelligence and 5G technology,the application research of advanced assisted driving system(ADAS)in intelligent driving is also accelerating.The intelligent sensor hardware platform and high-performance decision-making algorithm are important parts of the implementation of the ADAS system.In the automotive field,it is necessary to achieve vehicle-level standards and meet safety,reliability,and accuracy.This has higher requirements for the appropriate hardware and high-performance algorithms.Although millimeter-wave radar and lidar are widely used in the field of ADAS,they also have disadvantages such as high cost,lack of recognition ability,and less visual information.In view of this,this article mainly proposes an image ADAS algorithm based on the on-board system-on-chip(SOC),and designs and implements a complete vision-based ADAS system on the on-board chip.In order to enable the system to be effectively applied in actual scenarios,this article mainly conducts research from two aspects: algorithm and hardware.The image ADAS algorithm mainly includes three parts: Lane Departure Warning(LDW),Front Vehicle Collision Warning(FCW),and Traffic Sign Recognition(TSR).First,by studying machine vision image processing technology,analyzing and verifying common feature extraction and classification methods,this paper proposes a cumulative probability Hough line slope fitting lane line algorithm on a theoretical basis.After completing the preprocessing to enhance the features,according to the camera Set a specific ROI area in the position,use the canny edge detection algorithm and the cumulative probability Hough line slope twice to fit the left and right lane lines in the perspective space,and design the determination method of lane departure warning.At the same time,the HOG and SVM vehicle detection algorithm based on the parallel acceleration processing of the vision chip APEX and the SSD vehicle detection algorithm based on the fusion Mobile Net network are proposed,and the monocular distance measurement principle is used to realize the distance measurement of the target vehicle.For traffic sign recognition,a SVM classification speed limit sign recognition method based on edge super-pixel discriminant preprocessing is designed.In order to fully verify the proposed algorithm,through the comparison and analysis of domestic and foreign automotive chips,mainly considering the cost and computing resources,the environment perception chooses the automotive monocular camera module,which integrates the lens,AR0143 CMOS chip and image signal processor(ISP).Considering that processing visual information is processing each frame of video image,a lot of computing resources are required.The automotive vision microprocessor S32V234 has an image acceleration core APEX,which can quickly realize image and video target detection and recognition.Use S32 DS to compile and develop the tool to build the application program,and use on-board chip resources to verify the algorithm proposed in this article.With the breakthrough of deep learning technology,this paper also uses Texas Instruments’ TDA2 x vehicle chip to propose and design a vehicle detection algorithm based on deep learning.Through the verification on the on-board chip,the complete ADAS system is designed,and the experimental results show that the algorithm can be effectively applied and can meet the real-time performance in the actual scene,which provides a reference for the application of autonomous driving technology.