In order to correctly identify the poses of persons in an image, a technique called "human pose estimate" locates body key points. Human action recognition, sports, tracking, HCI, sign communications, and video surveillance all require this step to be achieved before moving on to the next stage of computer vision. The purpose of this article is to fill in the information vacuum and shed light on the studies of two dimensional human pose estimation. According to the number of persons to be tracked, it can be classified as a single-person or multi-person pose estimation. Afterwards, the many methods for determining a human's position are discussed and several uses and drawbacks are also mentioned. Using a Random Forest model, we address the challenge of 2D human pose estimation in still photos. We propose that picture patches be used to learn a human body's inception module. To extract E-HOG features and train a regression forest, we use patches randomly selected from a bounding box around a specific individual. It's also possible to estimate the joint density function's modes from aggregated leaf samples using an efficient technique. We use three publicly available datasets that include self-occlusion, appearance, and position variants to demonstrate this aspect of our holistic approach. A new dataset is also proposed, which differs from other datasets because of its different resolutions and distortion in the data. A better or similar outcome was reached when we compared our method to current best practices.

Human pose estimation, Random Forest, pose estimation and action recognition