Remarks on real-time human posture estimation from silhouette image using neural network

Abstract

This work proposes a human body posture estimation method from image, using ANN. 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 human body's significant points, such as head, shoulders, hands, elbows, knees, and feet. The proposed estimation method is implemented on a personal computer and runs in real-time. By using the estimated results, the estimated human posture is reconstructed with a stick figure in the real time application. Experimental results show both the feasibility and the effectiveness of the proposed method for estimating human body postures.