复杂背景下视频人脸的定位与跟踪

Video Human Face Location and Tracking in Complex Backgrounds

本文提出了一种视频人脸的定位与跟踪算法,包括人脸检测和人脸跟踪两个方面。该算法首先提取人脸训练样本的轮廓,抽取轮廓特征,用支持向量机对轮廓特征进行分类,实现人脸的检测;然后建立特征空间模型,用相似度函数选取最优的特征空间,简化跟踪难度,减少特征的数量,实现人脸的跟踪。实验表明,算法鲁棒性好、速度快,在背景与跟踪目标颜色非常接近的情况下,也能够准确地跟踪和定位出目标人脸。

This paper proposes a new method of detecting and tracking video-based human faces, which includes the aspects of human face detection and object tracking. First, in order to implement human face detection, the algorithm extracts the profile features of human face training samples and uses SVM to classify those features. Then, it establishes a feature space model, uses a similarity function to choose the optimal feature space, which simplifies the tracking difficulty and reduces the number of features. Experiments show that the method is robust and fast, even when the color of the background and the tracked objects are very similar, and it can accurately track and locate the object human face.