Font Size: a A A

The Research Of Moving Object Recognition And Tracking Based On Image Sequence

Posted on:2009-08-10Degree:MasterType:Thesis
Country:ChinaCandidate:C ZhangFull Text:PDF
GTID:2178360245979816Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Moving Object Recognition and Tracking Based on Image Sequence is the one of major research directions in computer vision. It is the basic and key technology in the application of moving robot navigation, intelligent surveillance, precisely guided weapon and human-machine interface. The algorithms of moving object detection, multi-objects cluster and object tracking are researched and a multi-objects recognition and tracking software system based on image sequence is realized in this paper.In order to detect moving object, the inter-frame and Horn-Schunck optical detection algorithm are analyzed. An optical detection algorithm based on HSI color model is presented according to the characters of dynamic background and color image sequence. The optical restriction equation in HSI color space is constructed and matrix reversion method is used to calculate optical vector instead of gray scale and iteration calculation. Thus it avoides information lost in color conversion and promotes the velocity and accuracy of optical calculation.According to the object character in color image sequence, a multi-objects ant-cluster recognition algorithm based on HSI color histogram feature is presented. The HSI color histogram feature of object is extracted. Ant structure and pheromone matrix is constructed. Ant update, pheromone matrix update and local search strategy are used to find the best solution which represents cluster result.Aiming at the deformation of object in tracking, a particle filter method based on self-adaptive ellipse deformation template is presented. The color-gradient histogram mixed feature is extracted and ellipse template based on affine transformation is constructed. The tracking process is realized by using particle filter based on Bayes Important Sample(BIS). Comparing to the single feature and fix template, our method solves the problem of tracking failure caused by color similar and object deformation. It promotes the accuracy and robustness of tracking.A multi-objects recognition and tracking software system based on image sequence is realized according to the research. The system is developed on VC.net platform with Direct show, multi-threads and GDI technology. It realizes the modules of image grabbing and preprocessing, multi-objects detection, cluster and tracking. The system can detect, cluster and track multi-objects in image sequence accurately. Experiment result shows that the system promoted efficiency and robustness in detection and tracking process. It also has a wide application perspective in complicated background occasion.
Keywords/Search Tags:Object detection, cluster, object tracking, HSI color space, optical algorithm, ant algorithm, particle filter
PDF Full Text Request
Related items