The Research Of Block-based Anti-occlusion Object Tracking Algorithm

Object tracking has always been the hotspot in the field of computer vision and can be applied to military,medical,video surveillance,virtual reality,and human-computer interaction.However,the current object tracking is faced with many problems,such as complex scenes(object occlusion,image sequence low resolution,complex background,object rapid movement,object deformation,etc.)will reduce the object tracking accuracy and affecting the use of results.Therefore,it is necessary to study the object tracking algorithm which is robust to these problems.Aiming at the object occlusion problem in object tracking,the object was tracked by blocking.Firstly,Kernel Correlation Filter(KCF)is used as the basic tracking algorithm for each block to optimize it.The confidence is calculated by using the tracking response graph to determine whether the object block is occluded and adaptively update the model according to the occlusion situation.At the same time,in order to reduce the influence of low confidence samples on the model,the model learning rate is adaptively changed according to the confidence.the specific location of the object is determined according to the confidence of each block in the tracking process.When the object is completely occluded,the detector for the whole object is used to re-detect the object.When the object is reappeared,the position of the object is set in time,and the speed and accuracy of the re-detection can be improved by the strategy of multi-particle detection.The robustness of the tracking algorithm to the tracking problem of strenuous motion,motion blur,deformation,occlusion and so on is tested by experimenting on the test data set OTB-100.The improvement of the tracking performance is determined by the model update strategy and the learning rate adaptive updating method,and the block method can improve the tracking effect to a large extent,and the robustness of the complex scene is improved greatly.The use of the re-detection mechanism of the block method can achieve long-term tracking purposes.The final algorithm has better tracking effect compared with other trackers.And because of the speed can be achieved in real-time,so it can be applied to the actual scene.