Research On Automatic Tracking Of Moving Objects Based On Video Image

Target tracking technology is one of the research hotspots of computer vision.It has considerable application prospects in practical production and life,such as intelligent alarm and automatic target tracking of monitoring,automatic target tracking of UAV,automatic tracking of moving cells under medical microscope,etc.The purpose of target tracking is to get the target position in video,but because of the interference caused by occlusion,illumination changing and scale changing,there is a challenge to obtain the target position.Faced with the challenge,many scholars have proposed a large number of tracking algorithms,which are mainly divided into two categories: generative and discriminant.Firstly,four generative tracking algorithms,Kalman filtering algorithm,mean shift algorithm,particle filter algorithm and optical flow algorithm,are sorted out,and an improved method is proposed to solve the problem of target loss in the case of occlusion.When occlusion occurs,the frame difference method is used to detect the target.The experiment proves that the improved algorithm can successfully track the tracking target after it appears.Then,in the research of discriminant tracking algorithm,the kernel correlation filtering algorithm is selected as the research object,and an improved algorithm is proposed,which is improved in three aspects: scale invariance,single feature and template updating.Experiments show that the improved algorithm is better than the kernel correlation filtering algorithm.The calculation accuracy has been improved,but the calculation speed has decreased a lot.Finally,two kinds of tracking algorithms are tested and compared in OTB50 video dataset.The experimental results show that the discriminant tracking algorithm outperforms the traditional generative tracking algorithm in tracking speed and accuracy,but the discriminant tracking method will lose certain tracking speed in order to achieve higher accuracy.