Human Activity Detection from Still Images using Deep Learning Techniques

Human activity detection is an active research topic now, the difficult problem of fine-grained activity detection is often ignored. This paper proposes a method to detect human activity from still images. Iterative detection of human activity in a scene is another tough and exciting area of computer vision research. In our day to day life, we have seen implementations of automated cars, speech recognition, and various machine learning models. Unlike action detection in videos that have spatio-temporal features, still images can't be considered similarly, making the problem more complex. The current work solely comprises activities that involve objects to reach a simple answer. Based on semantics, a complicated human activity is broken down into smaller components. The significance of each of these elements in action recognition is investigated in depth. This system is based on detecting an individual's action or behaviour with the help of a single frame (image). Activity detection consists of various tasks like object recognition, pose estimation, video action recognition, and image recognition. Since the current paper is focused only on actions that involve objects, a dataset with specified classes is created. Images for this dataset will be chosen from different sources. This study aims at the development of computational algorithms for activity detection in still images.