Research On Object Tracking Based On Particle Filter Under Complex Environment

Moving target tracking is an important research subject in the field of machine vision fields, with a wide range of applications in safety supervision, military guidance, intelligent transportation, medical diagnose, human-computer interaction, robot visual navigation and etc. The goal of target tracking is to make computer imitate human vision, detect and track the moving object in the video, provide information for further analysis and understanding. Although target tracking has been widely researched and a number of effective algorithms have been proposed, because object has shape deformation and complicated background, real-time and robust is still the hot topic in target tracking.In terms of object detection, some common approaches in the field of object detecting are introduced. Focused on the symmetric difference method and the edge detection method based the Sobel operator, and analyzed respective advantages and disadvantages through experimental results. Based on the analysis, a novel algorithm is proposed by integrated use of the advantages of the symmetric difference and the edge detection method. The algorithm can be accurate and reliable on detection of the moving target.For the target tracking problem, aim at the particle degradation phenomenon and lack of diversity on common re-sampling, a resample algorithm based on genetic evolution is proposed, which can effectively eliminate the impoverishment problem of standard particle filter, and keep the diversity of particle set.Under complex environment, it is difficult to track target successfully. To solve the problem, the paper propose a novel object targeting approach which fuses color and SURF (Speeded Up Robust Features) in the frame of particle filter. Using particle filter search the location of candidate object to improve the efficiency of tracking algorithm; employing extended SURF descriptors to set up the target model, SURF remain invariant for illumination, scale and affine. Add color to SURF vectors to form extended SURF descriptors, it not only maintains the characteristics of SURF, but also makes use of the image color information. The experimental results prove that the proposed method is real-time and robust in different scenes.