基于Adaboost的人脸跟踪模型

his paper proposed a fast video face tracking model of embedded platform, and the test results on the ARM7 processor is also given. This model increases two strong classifiers on the basis of the Adaboost model, which is regional detecting unit (RDU) and skin colour filtering unit (SCFU) respectively. With the previous detecting results, the RDU estimates whether the detecting area contains face, and whether the size of the detecting area is appropriate, only the detecting areas that are passed by RDU will be input to the SCFU. The SCFU does colour space transformation for every frame image, figuring out whether the detection area contains enough pixels of skin. And then, dose face detection with Adaboost model. Experiments show that the skin colour detection not only accelerated the face detection, but also greatly reduces the error rate of the Adaboost model. The requirements of computing speed of the SCFU is pretty low, but the detection speed is ideal, so it's very suitable for embedded system.