Research On Human Complex Behavior Recognition Method Based On Mobile Phone Sensor

The Development of pervasive computing, especially the popularization of micro-sensors in our daily life, makes human activity and behavior recognition gained more and more attention so that many institutions start to research activity recognition systems. Moreover, the growing number of smartphone is growing and the developing of hardware and software, provide pretty convenient conditions to record users’ activities and habits. Thus, human activity recognition system which based on smartphone sensors become a research hotspot. Then, we use some sensors on smartphones and the property of human activity to recognize and analyze daily life human activities. In detail, we have studied the following three aspects:First of all, we studied some simple human activities and analyzed the characteristics of simple human activities using the time domain features and frequency domain features of the smartphone’s sensors. We trained our model to make it can recognize human activities in real-time with rather high accuracy.Secondly, we recognized complex human activities based on our previous works on simple activities. Due to the diversity of complex activities, we devised a novel method that fused multiple data sources and the model in simple activity recognition. The method has overcome constrain of device orientation and position, which makes it more universal and could recognize complex activities under some specific circumstances.Finally, we tried to recognize complex activities under multiple specific circumstances. Due to the additional complexity of activities, we chose a hierarchical classifier which use multiple classifiers hierarchically to ensure the high accuracy. We used similar feature extraction and processing method in our previous work.