Research On Moving Object Tracking Based On Particle Filter

Moving target tracking in video sequences has many important applications, such as vision surveillance, image compressing, three-dimensional reconfiguration, robot technology, etc. The difficult points of object tracking include sudden moving of objects, sudden change of object and background, occlusion between objects and background, and the moving of cameras. Moving object tracking under complex background is mainly studied in this thesis. An object tracking method with good real-time performance, accuracy and robustness is proposed, and can be applied in the area of mobile robot technology. Video tracking technology, especially the basic theory of particle filter is discussed in this paper; advantages and disadvantages of theories that applied on object tracking are also analyzed. Object detection is very important for an online real-time tracking system. Accurate construction of object reference model relies on accurate object detection. Therefore, object detection technology is also a research facet in the thesis. Three research results in this thesis are as follows.Firstly, adaptive frame difference method is used to detect moving objects. We make use of movement compensation to conquer the influence of the movement of vidicon on robots. Mathematical morphology is used to eliminate noise, then, object is extracted precisely.Secondly, color histogram and edge histogram are combined to construct object's reference model, which could conquer the disadvantages validly brought by object model constructed by single character, promote the veracity of object tracking. The influence of illumination and deformation could be reduced by the quantification of color histogram and edge histogram proposed in this paper.Thirdly, we compute the Euclidean distance not only between particles and object color model, but also between particles and object edge model. The two distances were used as important evidence in particle weight updating. Synthesis weight is based on the weighting sum of color weight and edge weight. We can adjust the coefficients of certain weights based on the characteristics of background and objects, so as to adapt the changes of background and objects. The method has been implemented on MORCS-2. Experimental results demonstrate that the method in this paper is efficient and effective.