A Human Activity Recognition Method Based On Wireless Sensor Network And Data Fusion With Probability Estimate

Recent advances in sensor technology, wireless communications and micro electrical mechanical systems (MEMS) have enabled the development of wireless sensor network (WSN) on many fields such as ambient intelligence (AmI). In the applications of ambient intelligence, such as smart home and intelligent office, we could recognize the human activities based on wireless sensor network and provide a variety of intelligent services based on these activities. This problem has become one of the core issues among a large number of researches. However, as the complexity of human behavior, a single physical sensor always cannot be competent for the task of recognizing activity accurately. Then, one of the most problem in the research on smart spaces filed is that, how to study the data fusion with multiple sensor data to recognize the human activities. This paper will introduce a human activity recognition method, based on data fusion with probability estimates. With this method, it will classify the data from multi-sensors and get the probability estimates for each activity. Then, it fuse all probability estimates to one based on Dempster-Shafer evidence theory.The main works are as follows:1. First of all, a human activity recognition method based on data fusion with probability estimate will be introduced. This system gather the data from sensor nodes at the low layer, then send to the terminal through the gateway node. The operations including data preprocess and feature selection will be done on the terminal. Then, through the data classification and data fusion, the final result of activity recognition will be got.2. Analyze the application of data mining and data fusion in the system. This part is the core of the paper, the fusion method based on Support Vector Machine (SVM) and Dempster-Shafer evidence theory is the main content. With the right method, SVM can get the better result of probability estimate. And it is prepared for the next step of data fusion. In addition, in the part of D-S evidence theory, one improved method to realize the combination task will be adopted to reduce the time complexity.3. Finally, the experiments based on true data will be done to validate the effect of the activity recognition method. It is shown from the result that multi-sensors data can fuse to recognize the human activity more exactly, and this method can improve the efficiency and feasibility of the activity recognition system.