Research On The Key Technology Of Vehicle Assistant Driving Based On Computer Vision

Ensuring driving safety is always the fundamental purpose of automobile assisted driving system.In recent years,car ownership has increased rapidly,and traffic accidents have also increased.Improving driving safety is an urgent task.Big progress has been made in the fields of big data,machine learning,and computer hardware,providing a good platform for software and hardware.Computer vision technology is more dependent on The technology of hardware such as radar has a wide range of applications in various industries due to its low cost and similar information interaction with humans.Under this background,the car safety assisted driving technology based on computer vision has received comprehensive attention and is widely used in the assisted driving system of automobiles.Based on the current situation of traffic safety in China,this article uses computer vision technology to conduct research on traffic sign detection,lane line detection and pedestrian human pose estimation.The main research contents and results include:(1)Research on traffic sign recognition technology and deep learning training optimization methods.Taking advantage of the high accuracy and fast speed of the SSD model,the SSD model is optimized for the specific proportion of the traffic sign and the target is small,and a traffic sign detection model based on the SSD model is given.Experimental results prove that the model in this paper not only has good recognition of small targets,but also has a certain degree of real-time improvement;according to the characteristics of SGD method and Adam method,a hybrid optimization method based on SGD method and Adam method is given.(1)Experimental results show that the hybrid optimization method can reduce the number of iterations and make the loss function converge better.(2)The lane line detection technology is studied.The Sobel operator is sensitive to lateral edges and the Hsv model is more sensitive to yellow and white.A dual model based on the Sobel operator and the Hsv model is used to detect lane lines.At the same time,the sliding window search algorithm is based on actual needs.optimize.(2)The experimental results prove that the Sobel and Hsv dual models have good detection results on lane lines.After the sliding window search algorithm is optimized,the real-time performance of the lane line detection algorithm is greatly improved.(3)Pedestrian body pose estimation technology is studied.Using the high real-time characteristics of the Open Pose model,the Open Pose model is optimized,and a human pose estimation model for the Chinese human body is trained.(3)Experimental results show that the optimized model has improved the accuracy and real-time performance of Chinese human body pose estimation.(4)Writing of human body joint point labeling script.A script for modifying the joints ofhuman postures was written in Python language,which greatly increased the speed of establishing human posture data sets while ensuring accuracy.(5)Establishment of related data sets.Annotated a data set containing 60,000 domestic three key traffic signs as the data set for the study of traffic sign detection and lane line detection;annotated a data set containing 100,000 human joints of 20-45 year old Chinese as a text Study a dataset for pedestrian pose estimation.