Study On Face Tracking Algorithm Based On Improved Particle Filtering

Human face tracking is a very essential issue in computer vision, which plays an important role in face recognition, video surveillance, human-computer interaction, robot vision and so on. In recent years, researchers pay more and more attentions to human face tracking, which gradually becomes a popular research direction, because of computer hardware's performance enhanced greatly as well as price reduced. But how to overcome the problems caused by the background interference, the changes of facial expressions, the partial or total occlusion and the motions of camera makes it quite a challenging work. Human face tracking algorithm must satisfy the demands of the accuracy and real-time besides the robustness. So far, more than 10 thousand related papers have been published. They all try to find a practical method which can accord with the above three demands.Two tracking algorithms are designed in this dissertation to track the human face region and contour respectively.①A robust and real-time method based on improved particle filtering is adopted to satisfy the demands of robustness, accuracy and real-time. TheωPSOPF is introduced in this method to alleviate the problem of degeneration which always exists in the common particle filters. After the distribution of particles is optimized by theωPSO, particles are moved to the region of high likelihood. Sample impoverishment will not happen because there are lots of particles in the region of high likelihood and duplicating the very few particles with high weight many times can be avoided. AdaBoost classifier is used to initial the target tracking and update the target template. Updating the template and optimizing the distribution of particles can enhance the robustness and accuracy of the method. The weighted histogram which is calculated by interval sampling the pixels in the region of interest is used as the only clue to make the method faster. Finally, the proposed method is implemented on the platform of VC2005 with OpenCV. The experimental results verify the effectiveness of the method and its robustness to some interference.②The level set method proposed by Li is improved, so it can avoid the problem of re-initialization, be more suitable for the representation of human face contour and tracking. Then the improved level set method is incorporated with the framework of particle filtering and a'Combing Level Set Method and Particle Filtering for Face Contour Tracking'method is adopted. The face contour is represented by zero level set function and approached by evolution of the level set function in this method. The seven parameters related to the affine motion and level set function are seemed as the target state. The three energy functions involved in the revolution process are used as the target observation. Finally, the human face contour is tracked under the framework of the particle filtering. The method is implemented with Matlab and the experimental results prove the effectiveness of the method.