Research On Moving Target Tracking Method Based On Machine Vision

Machine vision is an important branch in the field of artificial intelligence.Target tracking technology is a significant research area in the field of machine vision,which includes image processing,applied mathematics and many other contents.At present,the target tracking technology based on machine vision has been widely used in the fields of coal mine safety monitoring,military application and so on,which is a frontier subject with important research significance and value.This thesis focuses on the research and improvement of single object detection and tracking methods in video sequences.The main works are as follows:Firstly,the detection methods of moving targets are studied.The advantages and disadvantages of the frame difference method,the optical flow method and the background difference method are analyzed by comparing the theory and experimental results of those methods.On this basis,the ViBe algorithm based on simple background difference method is emphatically researched.Considering that the Ghost region will be generated in traditional ViBe algorithm using single frame initialization,the three frame difference method is introduced to obtain the real background image to initialize the background model to eliminate the Ghost region.In addition,a method to judge the foreground pixels by using dynamic threshold instead of the fixed threshold is presented so that the problem that the inaccurate detection of the moving target caused by the background disturbance in the traditional ViBe algorithm can be solved.The threshold value can be adjusted adaptively according to the degree of changes in the background and the anti-interference ability of the dynamic changes of background can be enhanced.Then,experiments are carried out and the experimental results show that the moving targets can be detected accurately by the improved ViBe algorithm.Secondly,the tracking method based on the improved Mean Shift algorithm is studied.Considering that single color feature is used to describe the target by traditional Mean Shift algorithm,an improved Mean Shift target tracking algorithm based on HSV color feature and improved LTP texture feature fusion is presented.The algorithm transforms the original RGB color space into HSV color space,which is more adaptive for the illumination change,to describe the target.Furthermore,the rotation invariant mode and dynamic threshold calculation are introduced to improve the LTP operator.Then,the features can be extracted more effectively and the anti-jamming performance of the features to illumination change can be enhanced by the improved LTP operator.In addition,the reliability index is introduced in the two feature fusion to dynamically adjust the weight of the two features,so that the target can be described more accurately in the actual video scene by the features after the fusion.The experimental results show that the improved Mean Shift algorithm can achieve accurate tracking both in motion scenes that illumination changes and in motion scenes that the target color is similar to the background color.Finally,the tracking method combined with Mean Shift algorithm and particle filter algorithm is studied.Considering the advantages and disadvantages of the improved Mean Shift algorithm and particle filter algorithm,the two algorithms are fused and the improved ViBe algorithm is combined to detect the target,and then a multi feature fusion automatic tracking algorithm is formed.Automatic target detection instead of manual selection is used in the algorithm to reduce errors and increase the stability of moving target tracking.Experimental results show that the proposed tracking method can achieve fast and accurate tracking,strong real time and stabilization in the motion scene with moving target occlusion or in the motion scene with small moving target and fast moving speed even with a few particles.