Theory And Method For Multi-target Tracking On Moving Camera

Multi target tracking technology under the dynamic background is a difficult problem in video object tracking large, but "dynamic background" and "multi-target" is a very common scene. More and more researchers put a lot of energy to carry out research in this area. Dynamic multi-target tracking technology is widely used in many fields, such as intelligent video monitoring system, intelligent traffic monitoring system, military weapons aim accurately and visual guidance and tracking have multiple target tracking shadow.This paper presents a dynamic background detection of multi-target tracking algorithm framework. First the image is converted into gray image, then the image gray-scale equalization. Through the trained gradient histogram classifier to categorize, find the candidate objects in the images. Then train online template library, and match the candidate targets by sample database object. Through the experimental analysis, the feasibility of the algorithm is verified.In order to increase the robustness of the tracking algorithm, this paper joined the method of sparse optical flow tracking, multi-features can make the expression characteristics of target template better. The features used in the front not sparse optical flow tracking. Here we selecte100points to run the optical flow. Through the forward and backward flow, eliminat the large error point. From the experiment, this algorithm can handle new target appearance, target disappearance and the partial occlusion problem.Finally, the software used the algorithm in this paper are discussed in detail. The software is based on the MFC dialog box program and machine vision library. The software can track pedestrians in video effectively and stably and can provide a comprehensive process data and the final result.