Recognition Of Human Body And The Application In Pedestrians Counting System

Recognition of moving human body has been receiving increasing attention from researchers in the fields of image processing and computer vision recently. Recognition of moving human-body involves four main stages: (1) moving objects segmentation in a complex scene;(2) human body segmentation from moving objects;(3) to propose moving human body's feature vector;(4) recognition by classifier. As the base of recognition of moving human body, to propose the moving human body's feature vector is the key of the whole procedure. In this paper, based on the research on moving human body's types of characters in spatial domain and frequency domain, various characters are picked out: characters for boundaries such as Fourier Descriptors and radius of similar round objects, characters for regions such as the ratio of length to width, the ratio of space occupation and circularity, image's color and texture characters. These characters are screened by experiments in this paper. Finally, a decision-making vector is formed with effective characters to recognize.In this paper, various experiments are done to select appropriate artificial neural networks. Based on classified results, three layers error back propagation network is selected in this paper ultimately using Levenberg-Marquardt training algorithm to realize networks' classified function. Regularizing modulation is used to reinforce networks' adaptive ability.A Real-time Pedestrians Counting System is realized using the proposed algorithm in this paper based on the former research. The pedestrians counting system is a system which can count the number of pedestrians automatically. It provides exact data for the managers of markets to administrate. This system captures moving pedestrians in the gate of markets using professional device such as camera, computer and image capture card and so on. Moving human bodies in a complex scene are segmented in digital images using the knowledge of image processing. Finally moving human body's feature vector is used in this paper to recognize whether it is a human body and gives out thenumber of objects by artificial neural networks.The algorithm in this paper is validated by experiments and has more academic significance and practical value. Through extensive experiments, the proposed system exhibits the correct people counting results of over 98% in the environment of laboratory and 97% more or less in the environment of markets and schools which have more pedestrians. The experiment results demonstrate that this pedestrians counting system has preferable veracity and robustness.