Research On Human Motion Recognition Based On Pyroelectric Infrared Information

As the security problem drawing universal attention in the society, detection and recognition for human motion of intelligent monitoring system become a research focus. Human motion recognition has wide application prospect in the fields such as intelligent surveillance, perceptive interface, virtual reality, content retrieval and so on. The principle of human motion recognition based on pyroelectric infrared sensor (PIR) can be described as this: the infrared radiation by a motion human body involves feature information of human action morphology; this information can be detected effectively by the PIR sensor and exported in the form of analog signals. By analyzing and processing the analog signals, the feature information of different human motions can be extracted and used for the purpose of classification and recognition.In this thesis, PIR sensor with Fresnel lens arrays was used to detect infrared radiation of different motions, such as walk, run, jump, pick, kick and climb, and then the feature information of human motion was extracted from the PIR voltage signals and used for classification. Firstly an infrared data acquisition system was established in our research, and the PIR signals were collected by the system. Fast Fourier Transform was chose as the first method of feature extraction and the spectrum information was obtained. Wavelet Packet Analysis was used as the second method of feature extraction and both wavelet packet coefficients and energy spectrum were obtained from the decomposition in five layers. In the recognition process, Support Vector Machine (SVM) and K-means Cluster were employed separately and the recognition results from these two methods have been compared. At last, according to the characteristics of different motions, a hierarchical classification method was proposed based on the optimization of the feature extraction and recognition algorithm. The experiment results showed that the results of the hierarchical classification by extracting various features with different algorithms were much better than those of the classification by extracting single feature. The overall recognition rate was above 90%, where walk and run 96.67%, jump 86.78%, pick 84.31%, kick 89.25% and climb 89.85%.The research in this thesis demonstrated that human motion recognition could be achieved to some extent by analyzing the output signals of the PIR sensor, and this will provide a new technology in some low-security applications. Currently, research in the field of human motion recognition based on infrared information is still in its preliminary stage. The results of this thesis will promote the development of this technology to a certain degree.