| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 99 === Surveillance systems are to detect if there are people entering, leaving, loitering at the monitored areas. Automatic detection of such events is critical to the early response of security staffs in taking appropriate measures. Traditional surveillance systems are mainly image based, where background removal and foreground extraction is likely to be influence by the change of lighting conditions, rendering the detection of foreground blobs sometimes invalid and affecting the final event detection results. This paper proposes an image-depth-fused smart video surveillance approach, where additional depth information of the image, made available by depth camera, is utilized to better segment foreground blobs, making the detection of the events more reliable. In this framework, algorithms based on image-depth-fused information to detect each of the concerned surveillance events are designed and implemented. As a comparative study on image-based and image-depth-fused smart video surveillance systems, extensive experiments with different conditions of the monitored areas have been conducted and compared between the image-base Verint system and our implemented system. The results show that, within a range of four to five meters, our proposed image-depth-fused smart video surveillance outperforms the traditional image-based systems, especially under drastic change of lighting conditions. The proposed system is best suited for the surveillance at the entrance/exits of a building, stairways, and middle-size rooms and offices.
|
|---|
