Real-Time Human Body Posture Estimation Using Neural Networks.

Abstract

This paper proposes a real-time human body posture estimation method using ANNs. The network is composed of three ANNs and a decision logic unit. The ANNs’ input is the result of a function analysis on a human silhouette’s contour extracted from camera images and the ANNs’ output indicates the feature points’ positions on the contour. The decision logic unit synthesizes each of the ANNs’ output vectors and then the 2D coordinates of the human body’s feature points are calculated. The proposed method is implemented on a personal computer and runs in real-time (17-20 frames/sec). Experimental results confirm both the feasibility and the effectiveness of the proposed method for estimating human body postures. By applying the proposed estimation method to a stereo vision system, real-time 3D human body posture estimation can also be achieved.