Face detection and human motion estimation in image sequence

Abstract

For the tracking and detecting the number of faces, there are different approaches such as digital image processing or neural network. This project aims to develop a technique for locating face region using image processing, counting the number of people walking across the area, and the analysis of movement of people as recorded by a camera. The speed of computation must be fast enough for real-time application. Before starting the analysis, a chroma chart for the skin colour region in the CIE L*a*b* colour model is prepared. Afterwards, background subtraction is carried out to reduce the computational time and improve the accuracy for the subsequent operations. Pixels of the remaining areas, which are the potential face regions, are converted to 8-bit values representing the likelihood of skin colour. The skin regions can then be detected by an adaptive segmentation. Facial feature detection or shape analysis is carried out to verify whether the skin regions are faces or not. It is suggested to use shape analysis rather than facial feature detection because of the complexity of the latter approach. Some basic mathematical morphologic methods have been introduced for the preprocessing of the image. In the current stage, the system can locate all skin colour regions of an image sequence and form a complete shape for face analysis.