Automatic detection of human faces in uncontrolled environments: Identification of direction and movement

This paper presents an application for automatic face detection on video streams from surveillance cameras in public or commercial places. In many situations it is useful to detect where to are people looking for, e.g. in exhibits, commercial mals, and public places in buildings. Our application is designed to work with surveillance cameras that are already available in those places, and do not imply an extra acquisition cost. The paper begins with a review of techniques used for face detection, a brief introduction to the OpenCV library, and of the requirements for the application. We then describe our approach, and present the main algorithms we used. We then perform an evaluation of the application, analyzing its performance throughout the processing, and analyze the accuracy of the recognition. We performed tests in real-time and off-line. The results are encouraging, and we identify limitations and improvements that can be introduced to decrease the error rate.