Image Sequence Of The Human Face Tracking Algorithm

With the development of information technology and the requirement of applications, face detection, tracking and other related technologies have attracted numerous researchers. Various methods of face detection and tracking appear and related productions come out continuously.This paper reviews the development of the face detection and tracking, and introduces some algorithms about face tracking in detail. The author has done the research works as follow:1. The face tracking based on statistical learning. The article introduces classical Kalman filter and particle filter based on sequential Monte Carlo method, and illustrate their advantage and disadvantage using both simulate data and test video.2. CamShift and mean shift algorithms are explored, The CamShift algorithm is studied more detailed. An algorithm which combined face detection and CamShift algorithms is designed to track face object automatically.3. A multi-face tracking algorithm is proposed. Firstly, the face objects are detected using a face detection algorithm. Then the face is tracked which is most similar to each previous tracking object, according to some principles proposed by the author and the prediction of kalman filter.The investigations show that, the particle filter is a better tracking algorithm as it can keep the target even in the state of non-linear or non-Gauss, but it has a high computation complexity, so it's not suitable for the application of multi-face tracking system directly. The CamShift has a low computation complexity, but the target object belong to the foreground should have features that are obvious different from other objects or the background. In the case of human face tracking, the most frequently used feature in the CamShift algorithm is the probability distribution of skin color. But in most cases, different people have similar skin color; therefore, it is difficult to handle the situation when a face object is occluded by the others.Being focused on the indoor application, the proposed multi-face tracking algorithm combines the fast face detection algorithm proposed by Viola-Jones and the Kalman filter. It can run in real-time for it has a low computation complexity and it can also handle the situation when the tracked face is occluded to a certain extent.