Research On Moving Object Detection,Tracking And 3D Positioning Algorithm Based On Omni-directional Vision System

Omni-directional vision can provide 360 degree range of visual information,which has important significance for the system,for the robot positioning,visual monitoring,automatic driving,map building and other fields.This paper designs a kind of omni-directional visual system,and on this basis,the moving target detection,tracking and three-dimensional positioning algorithm and other issues are studied,the main work is as follows:(1)The theoretical model and system design of the omni-directional vision system are described.The theoretical model includes the camera model,the binocular stereoscopic vision model and the non-parallel binocular vision model.Based on these,the omni-directional visual model which could catch stereoscopic information is designed.According to the model,the mechanical structure of the omni-directional vision system is designed,and the hardware and software system is built.The visual information acquisition experiment of the omni-directional visual system is completed.The ability of the omnidirectional vision system which can catch omni-directional visual information in all domains and stereo vision information in some area is verified by experiment.The feasibility of designed omni-directional vision theoretical model is verified by experiment too.(2)The principle of the typical moving target detection algorithm is compared,and the advantages and disadvantages of different algorithms are verified by the experimental results.Based on these,the algorithm combined with three-frame difference and background difference is proposed.The experimental results show that the proposed algorithm can accurately detect the moving target and eliminate the moving target shadow in the background difference detection result.Compared with the background difference algorithm,the proposed algorithm‘s effectiveness is improved by 30.97%.(3)The principle of the typical moving target tracking algorithm is analyzed.The principle of Kalman filter and the principle of contour extraction are described and experimented.After analyzing the advantages and disadvantages of the two algorithms,the algorithm combined with Kalman filter and contour extraction is proposed.The experimental results show that the proposed algorithm can track the moving target continuously in the range of omni-directional viewing angle.(4)The feature extraction and matching principle of SURF is analyzed,the improved feature extraction algorithm and feature point matching algorithm of moving object is designed.The designed algorithm can improve the accuracy of the feature point detection and the matching of feature points.Based on the matching results,the three-dimensional positioning of the moving target is studied and experimented.The final experimental results show that the system can detect the three-dimensional trajectory of the moving target,and the relative error of the measured depth is 13.16% at the distance of 1150 mm.The research shows that the omnidirectional visual system can obtain all-round visual information,and the stereoscopic information can be obtained in some viewpoints,and the moving target detection,tracking and three-dimensional positioning are realized,which has engineering application in prospect.