| With the continuous development and maturity of information technology, the application of computer vision technology is more and more popular in people’s life and biological recognition which achieved by computer vision technology has been significantly used in t our daily life. Compared with traditional identification technology, biological recognition technology has the advantage of uniqueness and stability, so more and more related scholars has put their interest in this work. Now biological recognition technology mainly includes:the iris recognition, fingerprint recognition, face recognition and the gait recognition, etc. The moving Human object in video sequences can be identified by computer vision technology through object detection image processing, feature extraction and feature fusion. The analysis and recognition of moving human body has important value and significance in many fields (artificial intelligence, human-computer interaction, virtual reality and so on) and sensitive security situations (such as bank, airport, military base).The purpose of this paper is to improve the performance of the human body identification in video sequences and make the corresponding contribution for realizing intelligent monitoring system.The key of completing body identification is the extraction of effective and reliable human features which include facial features, fingerprint features, iris features, gait characteristics, etc. The gait features which have the advantages of distance, non-contact, difficult to conceal and disguise is more suitable for intelligent monitoring system. In recent years, a large number of gait recognition algorithms have been put forward.As a general rule, the recognition of moving human body includes object extraction, motion features extraction and classification three parts. In the base of large classical algorithms, the paper put focus on features extraction and classification and put forward a new algorithm for recognition which has a better performance. The main work and achievements of this paper includes:1. Extracted better moving human motion features. There are many features can be extracted in the process of human moving. In theory, the identification rate can be100%if we extract all features of human, but this is not realistic. If only one feature is used in recognition, the result of recognition can not meet the requirements. Based on this, the paper selected area of regional and joint angles of limbs to represent human movement, which both ensured the real-time and improved recognition rate. Human body contour is unique. In traditional, people used actual contour points to describe body contour, but the paper carried on improvement by dividing body area into several small regions and calculating area of every region to describe contour, which not only avoiding complex calculation of profile of detection but also reducing the effects of the false contour points to recognition rate. As human joints contain rich movement information, the paper extracted nine body angles to fuse with area of region for final identification.2. Research on classification and recognition algorithm. The features are service for identification. Nearest neighbor fuzzy classifier which is a kind of multiple classifiers fused by parallel structure is used for object classification. According to the differences of feature memberships, add weights to every feature for fusing and use the joint feature vector to make classification decision.Experiment results show that the proposed algorithm has better recognition performance. |
PDF Full Text Request