Human Tracking using Augmented Feature Propagation for Multiple Cameras with Non-overlapping Views

• Main Authors: Chen, Ke-Yin, 陳科引
• Other Authors: Huang, Chung-Lin
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/27528348497112995914

碩士 === 國立清華大學 === 電機工程學系 === 99 === Due to tremendous amount of crime activities have occurred recently, Security has become the important issue. Surveillance systems have been installed in home, airports, railway stations, department stores and other places. The traditional surveillance systems have required high human cost but with low efficiency. Therefore, new types of multi-camera surveillance system can automatically detect and continuously track the moving objects based on computer vision technology. Under some circumstances, the tracking of human objects may fail because of light change, unusual behaviors, clothes change between cameras, or staying in the blind region for a long time. It will generate path discontinuity. In this thesis, we make use of the relaxed features matching to solve the problem of missing object tracking. Furthermore, because of the viewing angle of the cameras or the objects’ moving directions are different, the captured features are not the same. We propose a concept indicating that the feature can be propagated between different scenes. The augmented feature can be used for cascading the objects paths. We propose Augmented Feature to correct the path by using the similar appearance of the objects across multiple cameras.