A Monadic Framework for Real-time 2D Multi Person Pose Detection for captured Images and Videos

In Computer Vision Field, the Real-time 2D Multi Person Human Pose Estimation (MPHPE) plays a vital role in empowering machines to have a perception of persons in captured videos. In this work, we introduce a framework for 2D multiperson pose detector. The proposed method utilizes non- parametric depiction that we termed as Part-Affinity- Fields (PAF) machine can learn these fields to connect body parts of persons in an image. It uses bottom-up approach system to accomplish high precision with real-time performance, in spite of number of individuals being tracked in an image. In earlier methods, in training stage the PAFs and part detection were refined concurrently. We show that only PAF enhancement is pretty much better than enhancement of both PAF & body part location brings results in a significant optimization in both execution time and precision. Not only has that it also kept the accuracy of each body part individually. This paper also reports the comparison of results along with the method with earlier methods. This method is also suitable for real time applications that can be run on embedded system with improved speed.