Research On Object Tracking Algorithm In The Video

Video from the visual image to reflect the objective things, through the vivid description ofthings, intuitive and specifically expressed the information elements. In today’s informationindustry in Japan, along with communication, network, the development of computer technologyand microelectronic technology, video technology has been applied in various fields, so theresearch of intelligent video surveillance system is particularly important. Video moving objecttracking algorithm is based on the analysis of each frame in the video image information, datamining, learning objective behavior and make a lot of motion capture, the information after a seriesof processing, and mark the tracking target in the image of the corresponding position. At the sametime, because of the complexity of the target environment conditions, and by tracking object itselfexists a variety of reasons, has resulted in a motion video target tracking algorithm research inmany aspects there are certain challenges.In this paper by studying the target tracking process, the tracking algorithm to study theinfluential factors, trying to establish a precise and efficient target model, by combiningclassification or state deduction and other methods, the image of the target and the background, toachieve the goal of fast track the target in video.In view of the current most tracking algorithm is using the target model and the characteristicsof target matching appearance, in the process of target tracking for a long time, the scale of thetarget and shape change, combined with computer visual error, put forward an efficient targetmodel is used to improve the efficiency of the track and the success rate. Segmentation afterextraction of target feature is used to modeling said the exterior structure, by using the method ofimage segmentation, this will be tracking the target area divided into multiple pixel block,combined with the feature of SIFT, form words, and calculate the weight of each word in thevocabulary book, as the appearance of the target model. Appearance model is used to determinethe target key points of the place, by using the pyramid Lucas-Kanade tracker to predict thepositions of the key point in the next frame, and mobile tracking window position. Pointdisplacement weighted effective to overcome the problems caused by target dimension and shapechange. The experimental results show that the target under the condition of deformation orillumination changes, algorithm can accurately and real-time tracking to the target.Due to particle filter framework tracking target tracking the search range is small, in this paper, a combined with particle filter framework and more examples of tracking algorithm, and keeptrack the paths to search at the same time, in the process of tracking is calculated from the firstframe to the current frame the best route, using state variables with color histogram feature, whenselecting a minimal solution, combined with color features and tracking algorithm, effectivelylimiting the target particle weights. At the time of calculating the particle weight, also uses themethod of the shortest path on the calculated when the target state of particle samples to a frame oftarget particles in path length, to limit the particle selection area, ensure the robustness andeffectiveness of the algorithm.