Vision-based human body posture recognition using support vector machines

This paper proposes a vision-based human posture recognition method using a support vector machine (SVM) classifier. Recognition of four main body postures is considered in this paper, and they are standing, bending, sitting, and lying postures. First of all, two cameras are used to capture two sets of image sequences at the same time. After capturing the image sequences, a RGB-based moving object segmentation algorithm is used to distinguish the human body from background. Two complete and corresponding silhouettes of the human body are obtained. The Discrete Fourier Transform (DFT) coefficients and length-width ratio are calculated from horizontal and vertical projections of each silhouette. Finally, these features are fed to a Gaussian-kernel-based SVM to recognize postures. Experimental results show that the proposed method achieves a high recognition rate.