Research On Vehicle Road Condition Perception Technology Based On Machine Vision

In recent years,with the rapid development of China's automobile industry,the level of automobile intelligence has gradually improved,and automatic driving has gradually become the future development direction.In this paper,based on machine vision,two key technologies of image target detection and image depth estimation are studied.A real-time road condition sensing system is constructed to detect and estimate the road target in the process of driving.First of all,the basic theories of automatic differentiator,convolution forward and backward calculation,deconvolution and up and down sampling are studied and verified by simulation.For target detection,an image target detection algorithm based on multi-scale feature extraction and distributed prior frame is proposed.In this algorithm,multi-scale features improve the ability of small target detection,and the setting of prior box reduces the calculation load.By comparing the mean average precision and positioning precision with other algorithms,the advantages and disadvantages of the algorithm are analyzed.Secondly,for image depth estimation,a self-supervised depth estimation algorithm for binocular image training and monocular image detection is proposed.The algorithm does not need to manually label the data set,which effectively avoids the problem of unbalanced output caused by binocular training during monocular image detection.The image depth estimation algorithm compares with other algorithms through indicators and analyzes the advantages of the algorithm.Finally,the self-made data set is used to simulate the above two algorithms in day,night,rainy and other scenes.The experimental results show that compared with other methods,the target detection algorithm and image depth estimation algorithm in this paper have certain results on the above indicators Improve,basically meet the requirements of autonomous driving system for road condition perception.