CAJ | 학술논문

当前的人脸特征点定位跟踪方法因其计算量大，实时特性欠佳。给出了一种基于改进Viola-Jones算法和Kalman滤波器预测机制的定位及跟踪算法。该算法通过使用改进的Viola-Jones算法对本次人脸特征点进行定位，同时使用Kalman滤波算法对特征点下次出现位置进行预测，缩小了下一帧特征点定位过程中特征点的搜索范围，因而缩短了定位搜索时间。实验结果表明该方法在保证定位准确性和鲁棒性的同时明显增强了算法的实时性。
당전적인검특정점정위근종방법인기계산량대，실시특성흠가。급출료일충기우개진Viola-Jones산법화Kalman려파기예측궤제적정위급근종산법。해산법통과사용개진적Viola-Jones산법대본차인검특정점진행정위，동시사용Kalman려파산법대특정점하차출현위치진행예측，축소료하일정특정점정위과정중특정점적수색범위，인이축단료정위수색시간。실험결과표명해방법재보증정위준학성화로봉성적동시명현증강료산법적실시성。
The current research of location and tracking methods for facial feature point are poor in real-time performance because they are large in computing capacity. In this paper, an improved method based on Viola-Jones algorithm with Kalman filter prediction mechanism is presented. The current facial feature point is located by using Viola-Jones algorithm and the scope where the next feature point will appear is predicted by Kalman filter algorithm. As a result, the scope of the feature point in next frame is reduced and the locating time is shortened. Experiments show with this method the real-time performance of facial feature point location and tracking algorithm can be improved apparently while ensuring the accuracy and robustness.