基于SURF特征点的多人脸跟踪方法研究

RESEARCH ON MULTI-FACE TRACKING METHOD BASED ON SURF FEATURE POINTS

针对视频序列中多目标人脸跟踪问题,提出一种基于SURF(Speed-Up Robust Features)特征和KLT(Kanade-Lucas-Tomasi)匹配算法相结合的特征点跟踪方法。通过融合该方法,创新性地设计了一种多人脸跟踪系统框架,在目标出现明显的姿态、尺寸变化,或者遭遇局部遮挡、光照不充分等复杂环境干扰下,能够实现对目标人脸稳定跟踪。通过多组实验数据的对比,证明了该跟踪方法比Mean shift算法、传统KLT算法具有更好的鲁棒性,能获得更精确的运动信息;验证了多人脸跟踪系统能够在复杂环境下实现对多人脸的有效跟踪。

Aiming at the problem of multi-object face tracking in the video sequence, we propose a feature point tracking method which combines the SURF feature with KLT matching algorithm. By fusing the proposed method, we design a frame of multi-face tracking system innovatively, it is able to achieve stable tracking on the object face when the object appears obvious changes in gesture and size, or encounters the interferences from complex environments of local shelter and inadequate lighting, etc. By the comparisons of multiple sets of experimental data, it is proved that the proposed algorithm, compared with mean shift algorithm and traditional KLT algorithm, has a superior robustness and obtains more accurate motion information. In addition, it also demonstrates that the multiple faces tracking system can achieve effective tracking for multiple faces in a complex environment.