A Review of Activity Detection Methods Used in Videos Streaming

Abstract

Activity detection in videos embraces many recent sub-research directions, such as Action Recognition (what type of activity is being performed), Temporal Action Detection (that states the time of action occurred in a large video), Spatio-Temporal Action Detection (localizing activities both in space and time). New advances in convolutional Neural Network Architectures and increased computing resources have made it possible. 3 Dimensional CNNs outperform 2D CNNs in balancing both spatial and temporal information in activity recognition while working with videos. Various approaches which incorporate these networks have been discussed in the paper. Locating a specific action in a video is an advanced and more complex task. Our focus of the paper is to give a summary of methods and advances used in the domain of action recognition and action localization.