Design And Implementation Of Lane Line And Vehicle Recognition System Based On Jetson TX2

With the development of China's economy and the improvement of people's living standards,automobiles have become a necessary means of transportation for daily life.The increase in car ownership has led to a growing problem of traffic.How to effectively avoid traffic accidents has become a problem that is closely related to our lives.Therefore,it is of great significance to study vehicle and lane recognition.This paper introduced the research background and research significance of vehicle and lane recognition system,analyzed the research status of related fields around the world,and explained the research content and technical index requirements of the embedded recognition system.Based on the principle of vehicle and lane recognition,we proposed the recognition algorithm used in this paper.The lane detection algorithm based on edge feature and perspective transformation,and we introduced the implementation of lane detection algorithm from image preprocessing,edge extraction and lane recognition.The vehicle recognition algorithm based on improved SSD network,and we introduced the implementation of vehicle recognition algorithm from the establishment of the data set,change the base network of SSD to MobileNet.Then the hardware design of the vehicle and lane recognition system realized in this paper is introduced,and the overall architecture of the hardware is given.Then,we introduced the power supply design of each module of the system,the design of HDMI to MIPI circuit.We introduced the selection of image processing module and part of the peripheral auxiliary circuit design.We also introduced the ECU control module design.The software part is introduced from the framework of the software.The chip configuration program of the input module,the SBC chip driver and the vehicle recognition program and lane detection program on the Jetson TX2.We finally tested the performance of the entire system.The hardware performance is good,the system hardware can run normally,and the data communication is stable and normal.After testing,the accuracy of the lane detection designed in this paper reaches 92.7%,the detection time of single frame image on the platform of this system is 28 ms.The AP of the vehicle recognition algorithm on the self-built dataset has reached 82.1%,and the recognition time of single frame image is 0.14279s.Our System meets the design requirements in terms of accuracy and real-time performance.