ردیابی یک چهره شناسایی شده با برنامه ریزی پویا

In this paper, we consider the problem of tracking a moving human face in front of a video camera in real-time for a model-based coding application. The 3D head tracking in a MBC system could be implemented sequentially as 2D location tracking, coarse 3D orientation estimation and accurate 3D motion estimation. This work focuses on the 2D location tracking of one face object through continuously using a face detector. The face detection scheme is based on a boosted cascade of simple Haar-like feature classifiers. Although such a detector demonstrated rapid processing speed, high detection rate can only be achieved for rather strictly near front faces. This introduces the ‘loss of tracking’ problem when used in 2D tracking. This paper suggests an easy method of solving the pose problem by using the technique of Dynamic Programming. The Haar-like facial features used in the 2D face detector are spatially arranged into a 1D deformable face graph and the Dynamic Programming matching is used to handle the ‘loss of track’ problem. Dynamic Programming matches the deformed version of the face graph extracted from a rotated face with the template taken online before ‘loss of tracking’ happens. Since the deformable face graph covers a big pose variation, the developed technique is robust in tracking rotated faces. Embedding Haar-like facial features into a deformable face graph is the key feature of our tracking scheme. A real time tracking system based on this technique has been set up and tested. Encouraging results have been got and are reported.

We consider the problem of detecting and tracking an arbitrarily moving human face in front of a video camera in real-time. Real-time face tracking is a key step in many real-world applications such as Model-based coding (MBC), teleconference, human-machine interface, face animation, visual surveillance, etc. The targeted application in this work is MBC. The goal of face tracking in MBC is to extract motion parameters of the 3D head motion [1].

Through empirical evaluation, we estimated that one near frontal face template could cover around 60° of rotation around any axis with rather good DP matching result, a huge improvement of the original face detector. The advantages of our proposed scheme are listed below: • The system only makes use of an existing trained face detector, it does not require re-training or training on any specific target face. All the system needs is the knowledge from an existing face detector. • The system takes one template face graph whenever a ‘loss of track’ event happens, from a previous successful tracking. Thus the template face graph updated automatically. • The system can bootstrap itself and thus operate in a completely automatic manner. Our experiments show that a combination of the Haar-featurebased object detector with a DP module could lead to an easy and simple solution to fast and robust object detection and tracking. Our system could be considered as an accessory to work of a Haar-like feature based face detector. Although we only experiment with a special object, the human face, we believe that it is applicable to more general cases such as object detection and tracking problems.