Research On Vehicle Target Tracking Method Based On Vehicle Video

Vehicle tracking,as a research hotspot in the field of computer vision,plays an important role in unmanned,intelligent vehicles and assisted driving systems.However,there are some problems such as low tracking success rate and inaccurate positioning,especially in the case of complex scenes.This thesis is based on the particle filter framework.The appearance model is established based on the fusion feature to describe the vehicle target and the particles are resampled selectively,which improves the accuracy of vehicle tracking under complex environmental conditions.Firstly,this thesis introduces the background and significance of the research.And the technical difficulties of vehicle tracking are summarized according to the research status at home and abroad.Meanwhile,the particle filter theory is introduced in detail.Secondly,the state space model is established based on the particle filter theory.In terms of the state transition model,the weighted average speed of nearest several frames is taken as the speed of current time,considering the variation of vehicle speed.And the approximate uniform motion model is considered as the state transition model,which makes the prediction result more accurate and reliable.In terms of the observation model,this thesis combines the perception features with that extracted by the principal component analysis method to adapt to the vehicle tracking problem.Then similarity is measured by the fusion feature in the tracking framework of particle filter,so as to update weight of the particle.Thirdly,the particles are selectively resampled in order to avoid degeneration.The motion of vehicle is taken approximately as the smooth movement.A small part of new particles are generated by prediction.And the others are resampled in the usual way.Finally,we verified the algorithm of this thesis utilizing a series of video sequences,which are challenging and public,and it was compared with other tracking methods.In addition,the video sequence of expressway recorded by automobile data recorder was further validated.The experimental results show that the algorithm of this thesis is robust and can track the vehicle accurately.