Active Contour Tracking Combined With The Target Movement Information

The target tracking technique is to obtain the trajectory of the target in the continuous sequence image with time through the processing of the sequence image in the video.Target tracking plays an important role in the field of computer vision,and is widely used in video surveillance,human-computer interaction,automatic driving and other fields.Target contour tracking can clearly express the edge contour of the target,provide the target shape change information,so it is an important research direction in the field of target tracking.Target tracking technology has been studied for many years,but there are still some problems to be solved,such as interference caused by complex environments,target topology changes or occlusion,light changes and so on.Therefore,it is of great significance to study a high accuracy and high performance tracking method.This paper designs a method to combine the target movement information and the apparent features to achieve contour tracking,which overcomes the dependence on color or gradient features of traditonal tracking,and achieve a good tracking in the scene where the apparent features of the target are not obvious,but the relative movement is clear.The research work of this paper is as follows:1)Extract the color feature and texture feature of the target and the background with the super pixel as the basic unit,and construct the target/background model based on the apparent features.Considering that the target or background in the actual scene has many apparent modes(multiple colors or Texture),it is difficult to get the correct classification result by using two kinds of classifiers.In this paper,a confidence map calculation method based on local information is proposed,and combined with the confidence map obtained by SVM classifier to obtain more reliable confidence map based on the apparent features.2)Introduce the optical flow to discribe the movement information of the target,the optical flow field can represent the moving speed and direction of each pixel.The calculation result of the optical flow field may be motion blur,or contain non-target area,which may cause the extraction of the movenment feature inaccurate.To sovel this problem,this paper proposes an adaptive optical gray level adjustment method,in order to expand the gray value difference between the target and background region,making the movement characteristics more obvious.Firstly,based on the superpixel,the local and the global optical flow direction histogram are used to divide the target and the background region.Secondly,determine the gray value split threshold of the target and background area,dynamically adjust the optical gray level,and obtain the adjusted optical flow gray map.3)The adjusted optical flow gray map and the confidence map based on the apparent features are used as two attribute to train the decision tree.The decision tree adopts the ID3 decision tree algorithm,which is constructed according to the maximum information gain criterion.Using the decision tree to classify pixels in the region of interest,obtain the confidence map which combines movement information and apparent features,using the map to guide the level set function envolue to the edge of the target contour.Finally,through multiple sets of experiments,and compare to other algorithms,verify the advantages of our algorithm.