Human Pose Estimation Based on GAN and DCGAN Models

As an important application based on the field of computer vision, the various human movements and gestures can be reconstructed by detecting the key articulation points of the human body. It is mainly used in human behavior recognition, human-computer interaction, and attitude tracking. However, the current human pose estimation models have many challenges, such as difficulty detecting the non-typical articulation points of the human body and inaccurate locating of the extremities. They are prone to error or lack of information in complex situations. This paper proposed GAN and DCGAN models to tackle this issue, which can improve the accuracy of human posture prediction. This paper mainly focuses on the contribution of Generative Adversarial Network to the detection of human key articulation points, revising the original model posture and obtaining a model that is closer to the real posture of the human body. The experimental results demonstrated that the model's accuracy is improved to a certain extent after using the DCGAN model. Furthermore, we note that in most cases, the performance of the proposed model is superior to others.