Research Of Human Target Identification Based On Wireless Pyroelectric Infrared Sensors

Human target recognition technology is the hot spot of current research, some car manufacturers, universities, research institutions have begun a study of the human body detection and identification, and some achievements have been made. Currently,there are several methods of human identification,one of them is based on the characteristics of the movement, the other is based on the shape information, and another is based on the pedestrian model. The traditional methods are basically based on the video, which needs expensive equipment and of high computational complexity,while pyroelectric infrared sensors can detect infrared radiation emitted by human body around, which can be utilized to human movement detection within the effective range. It has a huge advantage of human detection and recognition because of its low cost, low power consumption and high sensitivity. However,using pyroelectric sensors to human detection also has its flaws, which will generate a lot of redundancy and error data, so that wo must design optimization algorithms to process these raw data.This paper proposed a new feature extraction and data integration methods of human detection and identification using pyroelectric infrared sensors.Under the premise of guaranteeing high recognition rate, we devised a better solution to the large amount of data generated by the pyroelectric sensor and error rate defects. The main work of this Thesis includes the establishment of a target model of the Fresnel fiber-optic modulation based the pyroelectric wireless infrared sensor, the signal processing of the target model, the target signal characteristics extraction, classifier design, the analysis of the experimental results.This paper first analyzes the research background and research status of human recognition, giving an overview of the traditional methods of human identification. Secondly, we established the target model while human targets are walking through the Fresnel fiber array modulated pyroelectric infrared sensor array and preliminary processing of the simulated signal model using the short-time Fourier transform (STFT). Then we extracted the signal characteristics base on the non-negative matrix factorization(NMF) algorithm and singular value decomposition(SVD), and improved sparse constraints NMF algorithm to further reduce the complexity of the characteristics of the signal. Then we researched the study of the structure of the BP neural network learning method, the standard BP algorithm, the BP network training step, and the lack of BP network, and research based on a the improved momentum-BP network algorithm. The network is very easy to construct, and widely used in practice,and no rules for input data, some theoretical studies foundation is very solid, so we using BP neural network as the identification classifier. Finally, according to the previous theoretical studies and signal model of human target,we devised a recognition method based on NMF and SVD feature extraction and BP neural network classification.Based on the experimental results the identification rate can reach92.4%, this proves the feasibility of the proposed method.