Study On Road Condition Perception Technology Based On Vehicle Vision System

With the continuous development of China's social economy,the number of vehicle ownership is expanding rapidly,but at the same time,the total number of traffic accidents is also gradually rising.Automobile intelligent assisted driving system is one of the effective means to improve road traffic safety.The visual environment perception algorithm is a key component of the intelligent assisted driving system.In recent years,valuable research results have appeared endlessly.In particular,object detection algorithms and segmentation algorithms based on convolutional neural networks have good performance in simple scenarios.But there are still some limitations in applying the algorithm directly to the actual complex traffic scene.This paper studies the practical application of driving assistance system in complex traffic scenarios,and proposes an algorithm framework for object detection and segmentation in the environment of vehicle vision system.The main work of the thesis is as follows:1.At present,the object detection framework of deep convolutional neural networks is divided into two research directions:based on region proposal and regression method for directly regressing the region from the Network.This paper compares and analyzes the characteristics and principles of two different methods and their representative algorithms,uses YOLO algorithm as a detection algorithm for vehicle target recognition in road scenes.,and uses the method of superimposing different resolution feature maps to improve the performance of the object detection algorithm.2.In the section of object segmentation,this paper analyzes and compares the performance of the existing object segmentation algorithms.The semantic segmentation algorithm based on FCN can effectively obtain the pixel-level label of road environment,and thus travelable area and lane line in the road can be segmented.3.In order to meet the hardware environment of vehicle vision system and the requirements for real-time processing of traffic information capabilities in assist driving tasks,this paper shares a feature extraction network in the two different tasks of detection and segmentation algorithm,and performs joint training as a framework,which can effectively reduce processing time,reduce computing and memory consumption.4.In order to verify the performance of the dete;tion and segmentation algorithm,this paper conducts training and testing experiments in the open dataset.At the same time,in order to adapt to the situation of road environment in China,this paper establishes a domestic road traffic dataset.The dataset contains 9,000 manually annotated pictures.It has certain application value and reference significance to the research of road environmental target detection and segmentation algorithm.In summary,based on the research of visual perception technology,this paper explores the application and optimization of object detection and segmentation algorithms based on convolutional neural network in complex traffic environment,providing new ideas for multi-task requirements under complex traffic environment.