Theory And Application Of Object Tracking Using Probabilistic Models And Stochastic Algorithms

With the growing of universal safety awareness, increasingly grim of global anti-terrorism situation, and the rapid development of computer technology, visual tracking has become a hot spot and core issue in the field of computer vision, and has a broad application prospects in the national defense and security, video surveillance, intelligent navigation, and virtual reality. Using probabilistic models and stochastic algorithms theory and methods to achieve target detection, identification and tracking is an important branch in the field of visual tracking due to its solid mathematical foundation, the practical application versatility and robustness, and it has attracted increasingly more attention of scholars at home and abroad and made a lot of significant progress. Based on the recent theoretical and progress of the visual tracking fields at home and abroad, this thesis focus on two problems in the field of visual tracking using probabilistic models and stochastic algorithms, the first one is the robust tracking of fast random moving target which has complex dynamic model, and the other one is the effecitive tracking of contour objects which have strong clutter background. This thesis also performs a servo control model using PTZ camera to achieve a stable real-time kinematic tracking. The main contents are as follows:In chapter1, the related study background and significance of the subject are expounded. On the basis of referring to domestic and international associated documents, system components, key technologies and the current application research situation of object tracking using probabilistic models and stochastic algorithms are investigated. The problems and challenges in the research and application of effecitive and consistent tracking of fast moving object are analyzed, and the main research content and research ideas of this study is proposed.In chapter2, the fundamental theory of visual object tracking using probabilistic models and stochastic algorithms are introduced and analyzed, and the affectness of stochastic dynamic propagation model and target likelihood probability model to the tracking performance are also analyzed in detail.In chapter3, the problems and challenges of tracking of fast moving object are analyzed in detail, and two algorithms in dealing with these problems are proposed. The first one is a novel particle filter called Motion-Adaptive Particle Filter (MAPF) to track fast moving objects that have complex dynamic movements,and the other algorithm called Critical state transfor particle filter also proposed in this chapter. Experimental results show that the proposed method is robust for tracking objects with complex dynamic movements, and in terms of affine transformation and occlusion. Compared to Continuously Adaptive Mean-shift (CAM-Shift), Standard Particle Filter (PF), Velocity-Adaptive Particle Filter (VAPF), and Memory-based Particle Filter (M-PF), the proposed tracker is superior for objects moving with a large random velocities and accelerations.In chapter4, the methods of contour object tracking using probabilistic models and stochastic algorithms and muti-contour objects tracking using a probabilistic exclusion principle are studied and analyzed in detail. In order to address the problem of robust tracking under cluster environment, a novel particle filter called Inner-Contour Particle Filter is proposed to track contour under complex background. The proposed algorithm first uses Sobel edge detector to detect the edge information along the normal line of the contour, and then samples the inner part of the normal line to get the local color information and combined with the edge information to construct a new normal line likehood. After that, all the inner color information are used to construct a global color histogram. Finally, the edge information, local color information and global color information are fused together as a new conservation likehood. Experimental results show that the proposed method is robust for tracking contours under complex background, and it is also computationally efficient and can run in real-time completely.In chapter5, a novel fingertip detection approach using Cubic Bezier Curve fitting based on Monte Carlo Sampling is proposed. We aimed to achieve a robust scheme that could stably detect the fingertips in giving images. To that end, the proposed method first segment the hand region in a cluttered environment using skin color segmentation in YCbCr color space and get the silhouette of the hand. And then a Monte Carlo Sampling method is used to estimate the best fitted Cubic Bezier curves to the sub contour points centered in each contour point. After that, the curvature of the middle point of the estimated Cubic Bezier curve is calculated as the curvature of the point in the contour. Giving the estimated curvatures, the local maximums of a cumulative curvature curve are detected as the candidate fingertips. Finally, geometry feature analysis including convex hull detection and convex defects detection is applied to eliminate the valleys among the candidate fingertips. Experiment results using images including different type of hands show that the proposed approach is robust to noise and can locate the fingertips’position precisely.In chapter6, the structure model and imaging mechanism of Pan-Tilt-Zoom (PTZ) camera are analyzed, and the effectness to the tracking algorithms of controlling the zoom action of camera while tracking is studied. In order to track fast moving object, contour object and hand gesture using PTZ camera, a servo-model of PTZ camera is introduced to the probabilistic models and stochastic algorithms. The experiment results show that the proposed algorithm is effctive and robust in tracking objects using PTZ camera.In chapter7, the major work of the study is summarized, and the conclusions and innovations of the study are elaborated. At the same time, future development is predicted in order to provide references for the further research on this project.