Temporal Action Detection: A Survey

In the field of computer vision, action recognition for video understanding has been studied for a long time. However, the videos used in action recognition are trimmed videos processed by professionals for well representing predefined actions. In recent, many people have been able to upload and watch real-world videos from the development of many media platforms. These platforms also make it easier to collect and access such untrimmed videos. As a result, for video understanding, research on temporal action detection on untrimmed videos has been actively studied recently, as well as research on action recognition on trimmed videos. Temporal action detection can be categorized into offline and online action detection, and many temporal action detection methods have been proposed in both fields over the last few years. In addition, due to the recent promising results of deep learning in computer vision, the performance of temporal action detection approaches has been remarkably improved. In this paper, we introduce deep learning-based temporal action detection methods that have recently attracted attention.