Remarks on Neural Network-Based Human Body Posture Estimation Using HOG Features

Abstract

This paper proposes a human body posture estimation method using a neural network. The input feature vector of the neural network is composed with the histograms of oriented gradients calculated from input image and the output vector of the neural network indicates the 2D coordinates of the human body's significant points, such as head, hands, elbows, knees, and feet. Experimental results show both the feasibility of the proposed method for estimating human body postures.