| Summary: | 碩士 === 長庚大學 === 電機工程研究所 === 93 === For public safety issues recently, the video surveillance systems have been widely used to record and monitor people’s behaviors in public territories. If there is an emergent event, the possible information can be investigated immediately. In a practical application, human are detected first in the surveillance system, but the variant light resources in the scene influence the monitoring quality. To assist a practical video surveillance system, a human face detection system with multi image techniques is proposed in this study. Light compensation, the skin color segmentation, and the region growing methods are performed to obtain the initial face candidates first. Then the lip and eye locations are detected by the color and morphology information. Finally, the face structure and geometric relation are employed to calculate the facial score which benefits incomplete face situations. After the faces are detected, the camera on the Pan-Tilt platform is able to rotate to display the face in the center of the screen immediately. Furthermore, the deep information is used in the camera control to make this system more diversified. From experiment results under normal environmental conditions, the highest detection rates of mouth, eyes and face are 96%, 88% and 94% respectively. This implies that this proposed system has the ability to achieve the face-tracking goal.
|
|---|
