Research And Implementation Of Lightweight System For Traffic Object Detection

As the number of motor vehicles and traffic accidents increases year by year,the traffic safety problem is increasingly serious.However,the assisted driving system is a good solution to solve this problem.Traffic object detection algorithm based on computer vision is an important part of the assisted driving system,and research on traffic object detection algorithm has far-reaching significance.At present,with the rapid development of deep learning technology,various target detection algorithms with high-performance GPU have high detection speed and accuracy,but it is difficult to implement on embedded platforms with limited resources.Under this premise,this paper studies and implements a low-power traffic object detection system for vehicle-mounted restricted environments.The main work includes the following aspects:1.This paper analyzes the feasibility of traffic object detection on the vehicle embedded platform,and determines to realize the low-power traffic object detection system from two aspects of algorithm and hardware.In terms of algorithm,simplify and optimize detection algorithms to reduce the amount of computation.In terms of hardware,use a heterogeneous computing systems to improve computing performance.2.Studies the object detection algorithm based on machine learning and the object detection algorithm based on deep learning,and compares their advantages and disadvantages of various algorithms.Select SSD(SingleShotMultiboxDetector)algorithm as the main research material.Due to the large amount of SSD algorithm parameters and high computational complexity,it is difficult to implement real-time detection on embedded platforms.This paper optimizes and simplifies the SSD algorithm under the guidance of MobilenetV1 and MobilenetV2,and designed the default box for pedestrian,then completes the model training and evaluation on the server.The comparison experiment proves that the idea of MobilenetV1 and MobilenetV2 can effectively streamline the SSD algorithm.3.Taking ARM development board and Inter neural computing stick as the main body,a low-power traffic object detection system based on embedded is designed and implemented.Then the simplified SSD algorithm is transplanted to the designed lowpower traffic object detection system,and simulation experiments and actual road tests are carried out.The results show that the system can basically complete the real-time detection of traffic objects in the vehicle-mounted restricted environment.