DEEP LEARNING BASED HUMAN POSE ESTIMATION USING OPENCV

Abstract

In vision-based human activity analysis, human pose estimation is an important study area. The goal of human pose estimation is to estimate the positions of the human articulation joints in 2D/3D space from photographs or movies. Because of the complication ofreal-world settings and a wide range of human stances, vision-based human poses. Estimation is a difficult task. Deep learning's rapid advancement has recently attracted a lot of attention. The simulation of the processing and reasoning capacities of the human brain has received a lot of attention. The visual system of humans. As a result, it is critical to continue to investigate. Deep learning techniques are used to estimate human pose based on imagery. a video-based 2D pose estimation approach that incorporates a multi-scale TCE module into the encoder-decoder network design to explore temporal consistency in videos explicitly. At the feature level, the TCE module uses the learnable offset field to capture the geometric transition between neighbouring frames. We further investigate multi-scale geometric changes at the feature level by incorporating the spatial pyramid into the TCE module, which results in even more performance gains.