Person tracking is considered an important application in the field of video surveillance. A top view camera provides a wide coverage of scene and better handling of occlusion as compared to frontal view. The proposed top view based person tracking method mainly contains four modules, namely BLOB detection, standardisation, size estimation and tracking. In blob detection, foreground is extracted using segmentation, statistical operation, connected component labelling and morphological operations. The top view is radial symmetric thus using this property, the extracted blob is transformed to upright position using geometric transformations. This effectively makes the blob/person rotation invariant in the scene. Basic shape based features including width, height and body ratio are measured for each blob. On the basis of these features, the algorithm effectively distinguishes a person, no person as well as merged person blob. A simple tracker for each blob is formerly created to maintain different parameters for example, blob name, blob width & height, blob x & y coordinates, blob time, blob history, blob size and blob ratio. The proposed algorithm along with the seven attractive/popular tracking methods have been tested on different test sequences. The experimental results show that the proposed algorithm significantly improves results by achieving true detection rate of 95%.
|Journal||Journal of Ambient Intelligence and Humanized Computing|
|Publication status||Accepted/In press - 2019|
- Computer vision
- Connected component labelling
- Geometric transformation
- Person tracking
- Video surveillance