A Dynamic Gesture Tracking Algorithm Based onthe Improved Tracking-Learning-Detection
ZHANG Yi, YAO Yuanyuan, LUO Yuan, ZHANG Tian
National Engineering Research & Development Center for Information Accessibility, Research Center of Intelligent System and Robot, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
An improved dynamic gesture tracking algorithm based on TLD(tracking-learning-detection) is proposed to solve the problem of tracking drift in dynamic gesture tracking when the gesture is occluded. After tracking the gesture target successfully, an occlusion window method is incorporated to determine the degree of the gesture occlusion. When the gesture is partially occluded, the TLD learner is utilized to solve the problem. When the gesture is seriously covered, the Kalman filter is added into the TLD tracker to estimate the area in the current frame where the gesture may exist. Then the search range is reduced, which improves the processing speed of the tracker. Meanwhile, the direction predictor based on Markov model is added into the TLD detector to reduce the detection range of the detector and enhance the discrimination ability for the similar gesture trajectory. The experiment shows that the improved TLD algorithm has strong robustness in different environments and can quickly and accurately track the dynamic gesture trajectory. The proposed algorithm also improves the tracking drift problem when the gesture is occluded.
[1] 李文生,解梅, 邓春健.基于多点手势识别的人机交互技术框架[J].计算机工程与设计,2011,32(6):2129-2133. Li W S, Xie M, Deng C J. Framework of human-computer interaction based on multi-point gesture recognition[J]. Computer Engineering and Design, 2011, 32(6):2129-2133.[2] Kalal Z, Mikolajczyk K, Matas J. Tracking-learning-detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7):1409-1422. [3] 卢鹏.基于单目视觉的手势检测与跟踪算法研究[D].厦门:厦门大学,2013. Lu P. Research on gesture detection and tracking algorithm based on monocular vision[D]. Xiamen:University of Xiamen, 2013.[4] 张俊辉.基于 TLD 框架的手势检测与跟踪系统的研究与实现[D].武汉:武汉理工大学,2013. Zhang J H. Research and implementation of gesture detection and tracking system based on TLD framework[D]. Wuhan:Wuhan University of Technology, 2013.[5] 刘艳丽,余贻鑫.大系统马尔可夫模型状态转移概率矩阵的快速形成方法[J].天津大学学报:自然科学与工程技术版,2013,46(9):791-798. Liu Y L, Yu Y X. An effective method to formulate state transition probability matrix of Markov model of large-scale system[J]. Journal of Tianjin University:Natural Science and Engineering Technology, 2013, 46(9):791-798.[6] Kalal Z, Matas J, Mikolajczyk K. P-N learning:Bootstrapping binary classifiers by structural constraints[C]//IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA:IEEE, 2010:49-56.[7] Gopalan R, Hong T, Shneier M, et al. A learning approach towards detection and tracking of lane markings[J]. IEEE Transactions on Intelligent Transportation Systems, 2012, 13(3):1088-1098. [8] 程立英,张丹,赵姝颖,等.一种基于 TLD 改进的视觉跟踪算法[J].科学技术与工程,2013,13(9):2382-2386. Chen L Y, Zhang D, Zhao S Y, et al. An algorithm of visual tracking based on tracking-learning-detection[J]. Science Technology and Engineering, 2013, 13(9):2382-2386.[9] Jiang C Y, Zhang Y A. Reduced-order Kalman filtering for state constrained linear systems[J]. Journal of Systems Engineering and Electronics, 2013, 24(4):674-682. [10] Ran C J, Deng Z L. Self-tuning weighted measurement fusion Kalman filter and its convergence[J]. Journal of Control Theory and Applications, 2010, 8(4):435-440.