Anti-occlusion Struck Tracking Algorithm Based On Point Matching Algorithm

Target tracking is an important research field in computer vision and image processing.It has been widely used in the fields of Vision SLAM,attitude detection,motion gaming,smart home appliances and so on.The problem of target tracking is actually to predict the relative displacement and the state of a specific target in a sequence of consecutive video frames,It is the problem of continuously estimating the relative displacement and the scale position change of the object in the initial frame.Struck is a Tracking by Detection algorithm,which has the characteristics of high realtime,but it is different from the traditional Tracking by Detection algorithm.The traditional Tracking by Detection algorithm uses the method of positive and negative sample classification,trains and updates the classifier on-line to predict the relative displacement of the target,but the Struck target tracking algorithm does not use the traditional method of positive and negative sample annotation to train the classifier,instead,the relative displacement of the tracked object can be obtained directly by predicting the classification function by means of structured learning.The tracking accuracy of Struck algorithm is better than that of general Tracking by Detection Algorithm.However,in the case of large-scale deformation and occlusion,the tracking effect of the Algorithm is obviously reduced.In the analysis of Struck's algorithm,the classifier constructed by the Algorithm will import error information when the tracked object deforms in large scale or when the object is occluded,so much so that the target tracking fails or the target tracking box drifts.In this paper,a new adaptive tracking algorithm with occlusion and scale change is proposed based on Struck tracking algorithm framework,The specific improvement of the Algorithm in this paper is to add the threshold of occlusion and to introduce the judgment mechanism of occlusion and deformation,and to stop updating the classifier when there is a large scale deformation or a large scale occlusion,the occlusion detection failure caused by the quick change of the target size is solved by scaling the size of the search sample,and then the new tracking target box is followed to correct the position of the target box in the current frame,using point matching and Struck's online training and learning results,the team was able to predict and keep track of the target after it was completely occluded.The experimental results show that the proposed algorithm framework is robust to the occluded target tracking.