Research On Intelligent Vehicle Target Detection Algorithm Based On Information Fusion

Reliable perception is the basis for safe driving of smart cars.The traffic situation in China is more complicated than in European and American countries.There are many people-vehicle traffic scenarios.Establishing a sensing scheme suitable for Chinese typical road scenarios can meet the high reliability requirements of autonomous driving.In this paper,millimeter wave radar is used to collect data on obstacles,and the obtained targets are effectively screened and filtered to determine effective targets.Then,the effective target is projected onto the image collected by the camera through coordinate transformation,and the ROI is established to complete the detection of the obstacle.The specific research work is as follows:Firstly,using radar to build experimental environment.The continental MMW radar ARS-408 was selected as an experimental method,and the MMW radar was installed and calibrated based on the principle of ranging,speed and azimuth of the MMW radar.According to the radar data protocol,the radar data generation and analysis are completed.Finally,the effective target detection area of the MMW radar is selected,the empty targets are filtered,and the invalid targets are filtered to complete the determination of the effective targets.Second,using camera to build experimental environment.Based on the yolov3 detection algorithm,the idea of a spatial pyramid is used to modify the yolov3 main network part to form a yolov3-bt target detection algorithm.Specifically,the yolov3-bt detection algorithm is used to improve the target detection accuracy,and the detection speed is accelerated by adjusting the resolution of the input picture,thereby obtaining the optimal matching value between the detection time and the detection accuracy.Finally,according to the experimental verifications,the improved yolov3-bt detection algorithm solves the problem of taking into account the real-time and detection accuracy of obstacle detection ahead.Thirdly,the research on typical working condition sensing algorithms based on information fusion in China.After completing the camera calibration,the MMW radar acquisition target is converted to the image,and the conversion between the MMW radar projection coordinate system,the camera projection coordinate system,the camera coordinate system,and the pixel coordinate system is completed in order to complete the spatial data fusion.At the same time,it analyzes the typical road conditions in China and determines the fusion recognition scheme.The final experiments show that:for the detection of obstacles ahead,the improved yolov3-bt target detection algorithm in this paper has a higher accuracy than yolov3’s target detection,in which pedestrian recognition is increased by 7%and traffic obstacle recognition is increased by 5%.The fusion scheme based on millimeter wave radar and camera in this paper is faster than the improved visual inspection scheme yolov3-bt,which increases the detection time by 17.8%.At the same time,for pedestrians crossing horizontally in typical road conditions in China,the detection algorithm of this solution is 135 frames earlier than the ordinary visual detection algorithm,and the corresponding detection time is 4.5s earlier,which effectively improves the safety of ADAS.