Comparison of Human Body Posture Estimation Method Using Silhouette Image

This paper presents human body posture estimation methods based on analysis of human silhouette and investigates their characteristics. Two types of estimation method are presented: one method is based on both the heuristically extraction method of estimating the significant points of human body and the contour analysis of the human silhouette, and the other is using artificial neural network. In the former, the 2D coordinates of the significant points are located by applying the heuristically extraction method to the human silhouette and the contour analysis of the human silhouette. In the latter, the input feature vector of the ANN is composed with the result of analyzing a human silhouette image and the output vector of the ANN indicates the 2D coordinates of the significant points. In both methods, the estimated results are optimized and tracked by using Kalman filter. Experimental results show both the feasibility and the characteristics of the presented methods for estimating human body postures.