Research On Human Daily Behavior Recognition Based On Motion Capture Sensors

In recent years, more and more researchers begin to study human behavior analysis and recognition, and its application range is expanding. In the field such as medical rehabilitation engineering, body feeling game, movie and cartoon creation, virtual reality, the professional sport analysis, human behavior analysis and recognition technology is more and more widely used and also has great value. Traditional mature human behavior recognition research is mostly based on the optical signal, through the analysis of the video or image sequence. This way of behavior recognition based on optical motion capture has obvious advantages, but fundamental problems at the same time, human motion signal collection is uncertainty.In this paper, we propose a new program used in human motion analysis and identification. The difference with the traditional motion capture technology is that, this article focuses considering portability of motion capture and analysis, accuracy and timeliness. The program is designed to identify a set of human behavior based on motion capture sensors. In our system, we use wearable motion capture system to get the body posture data. The whole system is based on the quantification of posture pattern recognition, which means that if we use the quantization attitude data as input data, we can still rebuild the body's posture. Behavior recognition, the method uses an integrated longest common subsequence as the kernel function of SVM, training and human daily motion sequence similarity comparison, in order to achieve common behavior classification. We use the longest common subsequence as kernel function of support vector machine, which is from the traditional single point in time based on the information completely different. It uses the time-series information contained in the action, which can design a method in space on the human body daily behavior classification scheme.The main contribution of this paper is carried out in two areas of innovation. The firstisafter the comparative analysis of the human body model commonly used method, select the geometry point line constraint-based human skeleton model, human action combined with the collected signals to achieve a more accurate three-dimensional reconstruction and posture analysis. Second, in the completion of daily behavior recognition based on support vector machine, in order to more reasonable to quantify an abstract action, we introduced the long est common subsequence as a support vector machine kernel function, at the same time, the paper design of experiments, and based on motion capture sensors collect experimental data, combined with matlab simulation platform for the identification conducted experiments to assess, demonstrate the feasibility of the signal acquisition operation and recognition algorithm by combining sensors used in daily behavior recognition, and can analyze the human behavior and identify promising research development and application to make a useful contribution.