Real-time Pose Estimation is a very crucial element in comprehending human poses. It has a significant role in many areas of our day-to-day life. In this work, we describe an instantaneous technique to determine the individual’s stance in a picture. Here, we use confidence maps, which tell us about how confident the keypoint is located at that position and also our approach learns to associate the body parts with a non-parametric representation that preserves both limb alignment and position information, the term for this is Part Affinity Fields (PAFs). The bottom-up approach performs well in real time with great accuracy. Previous studies found that across training, PAFs and body part localization estimation both enhanced concurrently. We show that there is a significant improvement in runtime performance and accuracy when PAF-only refinement is used instead of integrated optimization of the body component location and PAF.