A Novel Pose Recognition Algorithm based on Multi-Feature Point Fusion for Training Images

Abstract

Efficiently implementing accurate human pose estimation is one of the most fundamental and challenging tasks in computer vision, then, the novel pose recognition algorithm based on multi-feature point fusion for the training images is proposed in this research work. This study considers 3-step framework to achieve the goal of efficient estimation. Step 1: non-maximum suppression of the gradient magnitude is considered in the improved Canny algorithm, the direction of the gradient can be defined as one of the four regions, and the edge information will be collected. Step 2: The Gabor feature and Haar feature are combined together to achieve the fused feature. Step 3: the convolutional neural network is used for pose recognition. y analyzing the bottom layer of the convolution operation, it can be seen that the convolution operation can only linearly transform the input and the efficient estimation result can be obtained. The proposed model is tested on the collected database, and the pose recognition performance is tested.