People Counting Using Visual Face Recognition Based On Neutral Network

With the continuous promotion and use of intelligent video system in daily life, video-based people counting system plays an important role in various occasions. A limitation of defects in the traditional people counting technology is without walking direction selectivity and people counting is limited by the sensors of devices. As video-based people counting technology is more intelligent than other methods, it becomes one of the hot spots relating to many fields such as machine vision, pattern recognition, image processing, artificial intelligence and so on. However, there are still many problems to emerge in the current intelligent monitoring systems.Based on the analysis of the development status and application prospect of intelligent video-based people counting systems, this paper propose a new method of people counting using visual face recognition based on neutral network. In this paper, we study the video-based people counting of entrance and exit and collect videos by setting up a single camera in head scene. The faces with rich information are selected to use as objective of detection and tracking. Firstly, during the image preprocessing of face detection, a novel coupled-templates for candidate face regions extraction after skin segmentation is proposed to overcome the difficulty that face regions are largely sticky to similar skin backgrounds and exclude large numbers of non-face regions. Secondly, a well-trained convolutional neutral network (CNN) is employed to recognize the segmented candidate face regions so that the accuracy of face detection can be improved further. Finally face tracking based on face detection for people counting is performed and an approach of combining and comparing the horizontal coordinates of the target centroid is proposed to deal with face tracking loss, repeated-counting and other difficulties.The experimental results show that the proposed algorithm based on skin color segmentation and convolutional neural network can effectively improve the detection rate of face detection, which provides a good statistical basis for subsequent people counting. At the same time, the proposed algorithm of combining and comparing the horizontal coordinates of the target centroid also effectively solves the problem of unnecessary repeated counting.