Skip to Main Content
This paper proposes an object-based video recording system, which can track and record the behavior of the moving objects under multiple distributed cameras with non-overlapping views. The object relationship among cameras is trained by using batch-learning procedure, and all probability matrixes are updated constantly for the long-term monitoring. While tracking the object moving across different cameras, we apply the spatiotemporal and appearance model to compute the correspondence. The tracking may be lost due to light variation, unusual behavior, or color change of clothes. Therefore, after building the initial tracking path, we use the dynamic programming technique to check if there exists any the missing linkage in the tracking path and rebuild a new path. Moreover, Hidden Markov Models are applied to connect the missing path according the training data. The experimental results show the tracking paths for multiple people tracking.