Search Of Moving Object Tracking Based On TLD Framework

Object detection and tracking is an important research topic in computer vision,it has been widely concerned in the field of computer vision.In recent years,great progress has been made in the object detection and tracking algorithms.Tracking based on detection algorithms,tracking based on learning algorithms and tracking based on detection and learning algorithms have been proposed.The Tracking-Learning-Detection(TLD)long time tracking algorithm has been paid attention to because of its good tracking performance in the case of target occlusion and disappearance.The TLD tracking algorithm is divided into three parts: learning,detection and tracking.Although the algorithm has good robustness in the aspect of target occlusion and disappearance,there are still some problems,such as on tracking accuracy,real-time and automatic initialization,etc.In this thesis,improving the tracking performance of the TLD algorithm is taken as the starting point,and the TLD tracking algorithm based on the single moving target is studied and improved.In order to solve the problem of the initialization automatic of moving target,the detection algorithm of moving target is analyzed and studied in this thesis.And then the detection algorithm based on saliency map is studied.Since the moving objects in the video fame can be viewed as a part of significant or significant information,a new algorithm for detecting moving object based on saliency map and double difference method is proposed in this thesis.The algorithm uses only the first three frames after the moving target appears in video to obtain the position of target.Firstly we use the SLIC algorithm to divide the input image into N super pixels,and then N super pixels are used to calculate the saliency map based on color contrast.To locate the moving target better,the saliency map based on color contrast and difference map which generated by the double difference are combined.In order to make the algorithm suitable for the motion of the camera,the RANSAC algorithm is used to compensate the motion of two consecutive frames before the detection.The experimental results show that the proposed algorithm can detect the target better in the first three frames after the moving target appears.In order to solve the problem that the TLD algorithm is easy to lost in the case of object rotation and reduce the detection time,we improved original TLD algorithm.Thesis uses the target displacement in the previous frame to reduce the detection area and achieve the dynamic adjustment of the detection range.And on this basis a Knn matching witch fusion of the Median Flow algorithm is proposed to improve the tracking accuracy.In this algorithm,the last target box is used to reduce the matching range and then the Knn matching algorithm based on ORB feature points is used to locate the position of the target in the next frame.In order to suppress the drift in the matching process,when the number of matching points is too small or matching points concentrated in a certain range of the target,the results of the Median Flow algorithm is added to the matching point.The experimental results show that the proposed method can improve the tracking accuracy of the TLD algorithm and reduce the TLD detection time to a certain extent.Due to the complexity of the TLD algorithm,it cannot meet the requirements of the real-time in practical application.How to improve the real-time performance of the tracking algorithm is still a problem to be solved in the future.