Human action recognition using key point detection and machine learning

In the field of Computer vision detection human activity is primarily being quite challenged. There are various methods and approaches for this job one of the popular techniques is key point detection which identifies the external skeleton of a human body, and these points can be further used to recognize or classify poses. As this method looks quite difficult on its own, there are several libraries out there which can perform this key point detection. Google’s Media Pipe is one such efficient library which has different functionalities which include hand points, human body pose, human pupil detection, human face mesh identification and background segmentation. This library has been trained with 30,000 samples and then the model is converted to work with OpenCV without any need of deep learning architecture. The main motive is to use machine learning for classifying poses from the points which are generated by the library. The proposed approach uses the BLR Body Language Rule, which captures the angles of limbs which play a vital role in identifying a human action and forms a dataset out of these angles and uses ML to learn the patterns from it. This is a fully automatic process and can be adapted to various situations such that a broad variety of applications can be possible.