Research Of Motion Object Detection And Tracing Method Based On Mutil-features Fusion

Motion detection and tracking is an important work in visual surveillance field, and is widely used in intelligent video browsing, image coding, traffic management, banking supervision fields. By improving motion detection algorithms, in particular, solving the occlusion problem, the intelligent level of monitoring system is further improved, and the motion-objects tracking is effectively implemented.The paper mainly studies motion-object detection and tracking methods based on multi-feature fusion, at present, many motion-object detection and tracking methods is proposed, each method has its own shortcomings and inadequacies. The method is proposed in this paper combined Markov random field segmentation and tracking method of fragment features template, focus on solving the multi-object tracking problem under the static and dynamic occlusion, different light conditions, and has good experimental results.The result of motion object segmentation will have a critical impact on motion tracking, so it is important to choose a good segmentation method. In the area of moving object segmentation, the paper presents an adaptive weights method based on a regional MRF segmentation. By using the adjacent pixel region and spatial correlation, the method adaptive update the system parameters of the energy function and can more accurately segment the motion objects[1], Markov random field model based on Markov Random Field and Gibbs distribution equivalence has better results in the field of image segmentation.In the area of motion tracking, object feature template is established based on multi-features (such as the NMI feature, integral histogram features, etc.) ,the paper is using fragment features template combined with Kalman prediction and motion direction information to further improve the algorithm accuracy and to track, and compare with camshift and particle filtering methods commonly used in recent years, the experimental results show that the paper method is superior to the above methods in real-time, accuracy, etc.