基于颜色和头部轮廓的粒子滤波跟踪算法-应用数学专业论文.docxVIP

中山大学硕士学位论文：基于颜色和头部轮廓的粒子滤波跟踪算法Head—tracking 中山大学硕士学位论文：基于颜色和头部轮廓的粒子滤波跟踪算法 Head—tracking Based on Color and Contour with Particle Filter Major：Applied Mathematics Name：Gno．wei Sun Supervisor：Professor Dao—qing Dai Abstract Moving objects tracking is of the active research direction in the field of computer vision．People get the exoteric information by sense，and vision information is the main knowledge they get．Image sequences of moving object contain more information than static images，while object tracking is new research subject with the development of digital video frequency technology．It has important significance in science and engineering and potentially widespread applications in many domains， such smart surveillance，perceptual interface，motion analysis，virtual reality． In this paper，we research the head-tracking of single people in images sequences． The thesis discusses the general objects tracking problem under the frame of Bayesian estimation and the common algorithm of particle filter．The particle filter based Sequential Importance Sampling is deduced in detail．We present particle filter based visual tracking with color and shape cues fusion．Various tracking algorithms have been propose，however，most of them are based single image cue．It is clear that multiple cues fusion Can provide more reliable observation．In this paper，we use ellipse to represent the human head．Combining the knowledge of head contour and face characteristic，we inosculate two different models to represent Our object， is ellipse model based on contour error，the other is color histogram model based color information of inner obj ect area．Then on hand we can improve the correctness by precise orientation of edge，on the other hand，improve robustness of algorithm by rich information of inner object．When considering the space continuity and regularity of the moving object between neighborhood frames，we adaptive velocity model to describe moving object，then particle filter to track the object．