|
Video target tracking is a core issue in the field of computer vision, and has a wide range of applications in both civilian and military, such as intelligent monitoring, human-computer interaction, robot navigation, and guided weapons, in recent years, with the rapid development of information technology, the goal The track has attracted the attention of many researchers, has become a hot research problem. Although it has been proposed many effective video object tracking algorithm, but in the practical application of video target tracking is still facing many difficulties, such as the lighting changes, changes in object pose and nonlinear deformation as well as background noise and interference, etc. design robust video object tracking algorithm remains a challenging task. In this paper, the particle filter framework, difficult problems in video object tracking for target appearance model designed to carry out in-depth research, proposed multi-feature adaptive fusion target observation model representation. Adaptive weights adjusting algorithm proposed particle filter for target tracking based on multiple feature weights adjustable. Adaptive adjustment of each feature weights, this paper through the analysis of the distribution of the particle filter particles, frame-by-frame weights adaptive update algorithm design. This method based on the particle distribution of the current frame can be a good tradeoff of the reliability of each of the feature, thereby to adjust its weights, in large part, able to adapt the tracking complexity of the environment, to ensure the accuracy of the tracking, however, the The method is vulnerable to the impact of the current frame error. Take into account changes in the weights of the continuity in the time series, designed a weight tracking strategy, using a particle filter to track the feature weights, and combined with frame-by-frame adjustment algorithm, the double particle filter multi-feature fusion target tracking algorithm. This method not only to achieve a frame-by-frame adjustment of the feature weights according to the actual situation, to avoid tracking error caused by the weight mutation, to ensure a stable and reliable tracking results. In this paper, weight adjustment algorithm and tracking methods, provide a reliable basis for tracking accuracy and robustness. The test results show that the video data with different track conditions, this adaptive weight-based multi-feature fusion strategy made very significant improvements compared with existing fusion method.
|