Research On Moving Object Tracking In Video Sequences

Moving target tracking technology is an important research content in computer vision.At present,many scientific research institutions,universities and enterprises have done a lot of research in this field.This technology has been widely used in military UAV,medical surgery,video surveillance and intelligent transportation,although the performance of computer has been greatly improved,Various excellent algorithms have been proposed,and the real-time performance,accuracy and reliability of moving target tracking have been greatly improved.However,the complex and changeable environment will interfere with the accuracy of moving target tracking.At present,no algorithm can meet all scenes.In view of the shortcomings of the existing algorithms,based on Open CV open source visual library,this paper studies the common algorithms in moving object tracking.(1)Research on image preprocessing algorithm.The gray-scale,binarization and denoising in image preprocessing are analyzed and experimented.The RGB color space model and HSV color space model are analyzed in color space conversion.The dilation and erosion,open operation and close operation are analyzed and experimented in image morphological processing.The advantages and disadvantages of various methods are compared.(2)Research on the detection of moving objects.Firstly,the traditional algorithm is analyzed theoretically and then implemented.Experiments are carried out in multiple scenes.Based on the real-time,accuracy and detection results,the advantages and disadvantages of each algorithm in different scenes are summarized.Based on the deficiency of traditional detection algorithm,the improved algorithm is a fusion of frame difference method and hybrid Gaussian modeling(GMM)algorithm.The improved algorithm is used to experiment in simple scene,fast scene,multi-objective scene and moving target and background gray close scene respectively.The experimental results show that the improved algorithm can eliminate the ghost generated by the GMM algorithm and detect the moving target with high quality,and improve the accuracy and recall rate.(3)Research on moving object tracking.This paper classifies the commonly used moving target tracking technologies according to their tracking characteristics.Firstly,the traditional meanshift algorithm is analyzed.Aiming at the shortcomings of meanshift algorithm,CAMSHIFT algorithm,an adaptive improved algorithm of meanhift algorithm,is studied.CAMSHIFT algorithm will lead to tracking failure in light changing scenes and occlusion scenes,The common CAMSHIFT + Kalman fusion algorithm can effectively solve the occlusion problem,but the tracking drift will still occur in the scene of light change.This paper proposes orb-lbp feature matching and CAMSHIFT + Kalman fusion algorithm,the improved algorithm can correct the tracking drift and achieve effective tracking.Experiments are carried out in two scenes with occlusion and illumination transformation.Under the three evaluation indexes,the proposed algorithm is more accurate and robust.(4)The design and implementation of the tracking system of moving targets,the above algorithms are added to the system,which can realize the functions of image processing,moving object detection and moving target tracking.The detection and tracking effects are viewed and the robustness of the algorithm is verified.