| Summary: | 碩士 === 逢甲大學 === 自動控制工程學系 === 102 === This research is developed by using the existing dual-axis twin-CCD image servo shooting platform, to perform the real-time tasks of facial recognition and human head posture estimation by using the dynamic image processing technique.
At first, we utilize the camera calibration method to find out the intrinsic and extrinsic parameters of the camera, and calculate the spatial three-dimension coordinate through some kinds of coordinate-transformation and geometric relations on the stereo vision.
For the estimation of human head posture, in this research, we use the Haar-like feature based on Adaboost algorithm to capture the human head feature points including right-eye and left-eye, and also recognize the mouth color using YCrCb color space. Therefore, based on the relative position of these three points in the space, the head’s posture can be estimated. Finally, we construct a two-axis head motion simulation platform for experimental verification. This technique of platform construction and posture estimation will be applied to the automatic shooting robot system.
|
|---|
