Block Object Tracking Algorithm Combined With Target Global Information

As an important sub-area of computer vision,visual tracking has a very broad application prospect.Target tracking is the arbitrarily given target in the video sequence,determining the motion trajectory or position of the target in subsequent sequences of the video.In recent years,single-target tracking technology has achieved rapid development,especially the introduction of related filtering and deep learning theory.Single-target tracking technology can be practically applied in some simple scenarios.However,due to the interference of occlusion,deformation,scale and illumination changes,single-target tracking in complex scenes is still difficult to meet the practical requirements.Based on the analysis of the research progress of single target tracking algorithm at home and abroad,this paper studies the problem of template regeneration pollution degradation and the block-based method does not pay attention to the tracking target as the overall information of the complete body.main tasks as follows:(1)When the target has a large range of occlusion,the update of the model may introduce non-target features,causing the target tracking observation model to be degraded,resulting in tracking drift.To solve this problem,a new feedback update strategy is proposed.Firstly,the initial judgment is made by using the maximum response value of the current frame target overall filter,and then according to the ratio of each sub-block filter and the change of the previous frame,it is further determined whether the target has occlusion;and then different update mechanisms are adopted according to the occlusion condition: no occurrence When occluding,all filters are updated with a lower learning rate,otherwise the unoccluded sub-block filter is updated with a higher learning rate.(2)The existing block-based target tracking algorithm can weaken the influence of occlusion to a certain extent,but in the tracking process,the block-based algorithm ignores the tracked target as the global information owned by the complete body,and only according to the maximum response value of the sub-block.Target positioning,resulting in low tracking accuracy,in response to this problem,propose a target tracking algorithm combined with the target global information.Firstly,based on the analysis of the influence of different block modes on tracking accuracy and efficiency in the target area,the target area is divided into four sub-blocks.Then the occlusionjudgment and update strategy proposed in Chapter 3 is introduced to update the filter.Finally,the multi-child is merged.The block location information determines the final location of the target.(3)In order to verify the effectiveness of the method,a comparative experiment was performed on the most authoritative data set OTB(Object Tracking Benchmark).On the feedback update strategy algorithm,the experimental results show that the new feedback update strategy effectively alleviates the problem of template pollution degradation when the model is updated,especially when the target is occluded,the method adopting the strategy has a more reliable model.The experimental results show that the proposed method can effectively determine whether the target has occlusion,improve the success rate of target tracking,and greatly improve the tracking effect under occlusion.Therefore,when the occlusion is not present,the target position is determined by using the complete target itself template,and the framing multi-sub-block position information is used to predict the target position of the next frame.