Human Activities Recognition Using OpenCV and Deep Learning Techniques

Abstract

Protecting privacy from hidden video is an important social issue. We need a computer vision system (for example, a robot) that can recognize human activities and provide help in our daily lives, but at the same time, make sure that it does not record videos that may violate our privacy. This article describes the basic method for resolving such conflicting tasks: Recognizing human activity when only anonymous low-resolution video (eg 16x12) is used. We will introduce deep learning, CNN, and OpenCV Paradigm, whose concept is to teach the best image conversion set to create multiple low resolution (LR) / high resolution teaching video concepts from video. Our concept is studying various types of sub-pixel conversion optimized for activity classification, which allows classroom experts to take advantage of existing high-resolution videos (such as YouTube videos) by creating some LR training videos suitable for this problem. We have verified through experiments that OpenCV in the computer paradigm can benefit from activity recognition of extreme human activity videos.