SVM-based approach for human daily motion recognition

In this paper, a novel technique for human daily motion analysis and recognition is proposed. The technique is based on the use of inertial sensors, and integrates a longest common subsequences (LCSS) algorithm as the kernel function for support vector machines (SVM), which measures the similarity of human daily motion time-series. In our system, we use the wearable motion capture system to obtain the body posture data. The entire system is based on the quantized body posture pattern recognition, which means, we can still reconstruct the body's posture if we use the quantified posture data as the input data. At the same time, we first use LCSS as the kernel function of SVM for action recognition, which is quite different from the traditional time-based single point of information, or the windowing method. It does take advantage of the time-series information contained in the action, so can we design a method to classify the human daily motion under the posture space methods.