A Deep Learning Model for Investigation on Human Body Movements and Action

摘要

Recognizing body movements are vital for fields like virtual reality, human computer interaction, and safety monitoring. The research seeks to overcome the limitations of existing body movement recognition systems, which include individual human body recognition rather than multiple recognition and also fails when the body is partially visible. Since most of the real-world applications, like accident monitoring requires recognition of multiple people; at the same time, the existing methodology has reached an impasse. Classification algorithms are predominantly used to recognize various body movements and actions. Even Kinect, which is well known for recognizing human actions and movements, uses a random forest algorithm to track the movements. With recent technological advancements, researchers are keen to increase the accuracy of tracking the movements. New research methodologies seek to overcome the above limitations by using deep neural networks and many other machine learning algorithms. With the inspiration of the success of convolution neural networks (CNNs), the implementation explores usage of the mentioned deep learning algorithm in estimating human poses and recognizes corresponding actions. The idea is to build a scalable model that can be used for many applications and get accurate results as much as possible.