基于特征点的3D人脸姿态跟踪

针对视频序列中的人脸跟踪问题,提出一种单摄像头的人脸3D姿态跟踪方法。利用SIFT特征点匹配算法来得到可靠的帧间特征匹配。将前帧与所选的关键帧特征匹配信息及融入到对当前的姿态估计中,利用SIFT特征点匹配算法来得到可靠的帧间特征匹配。最后通过利用RANSAC随机选取特征点对,并用POSIT和最小化误差组合的3D投影方法以迭代的方式得到精确的当前帧人脸姿态估计。通过多组实验数据对比,表明了该算法在严重遮挡、头部摆动幅度较大、匹配点较少的复杂情况干扰下仍具有鲁棒性,并且解决了3D人脸跟踪的漂移问题,实现对目标人脸的稳定跟踪,对比以往2D跟踪算法在复杂环境下具有明显的改善。

3D face pose tracking based on feature matching

Aiming at the problem of object face tracking in the video sequence, an approach was proposed for 3D pose tracking with a single camera. SIFT (scale invariant feature transform) algorithm is also used, which can get reliable inter frame feature match. And the feature correspondence information from either previous frame or some selected key frame is fused into the current frame pose estimation. The pose estimation is a classical problem which can be formulated as that of minimizing an error metric based on collinearity in object. Finally, the current face pose estimation is obtained via 3D project from combination RANSAC and POSIT algorithm. Various set of experimental data shows the improvement of our algorithm over existing 2D matching algorithms especially in solving the pose drifting question when tracking agile motion, severe occlusion, drastic illumination change.