Scene clustering with Multiple Non-Calibrated Cameras(CV)

概要

論文の詳細を見る
We propose a new scene clustering method for multiple fixed cameras that spread widely over daily scenes. The cameras do not need to be calibrated or to overlap with each other. Our new method utilizes regions of interest (ROIs) in multi-view images extracted using our previously described method. The scene clustering method discriminates between all possible events by checking snapshots observed in the extracted ROIs in the multiple views of the cameras. The advantage of our method is that it only requires a set of video data, and does not need to know the camera locations. We implemented a preliminary surveillance camera system with 35 cameras, and experiments confirmed the utility of the proposed method.
社団法人電子情報通信学会の論文
2006-11-17