Human Pose Estimation Based on Domain of Interest and Part Affinity Field

Abstract

This paper proposes an effective method to detect the 2D pose estimation of multiple people in an image. Learn human joint points through convolutional network, obtain the attention domain of joint points, classify the attention domain of the joint points, and obtain part affinity field of the information at the same time; combine the attention domain and part of the affinity field information of the joint points, using a greedy algorithm Perform clustering of joint points to finally obtain the human post. Simulation experimental data show that the method in this paper has achieved satisfactory results in performance and accuracy.