IARG: Improved Actor Relation Graph Based Group Activity Recognition

Group Activity Recognition is to recognize and classify different actions or activities appearing in the video. The detailed description of human actions and group activities is essential information, which can be used in real-time CCTV video surveillance, health care, sports video analysis, etc. The existing methods, such as pose estimation based and graph network based group activity recognition can perform reasonable group activity understanding, however those models have bad performance on video with extreme brightness and contrast condition. This study proposes an improved actor relation graph based model (IARG) that mainly focused on group activity recognition by learning the pair-wise actor appearance similarity and actor positions. We propose to use Normalized cross-correlation (NCC) and the sum of absolute differences (SAD) to calculate the pair-wise appearance similarity and build the actor relationship graph to allow the graph convolution network to learn how to classify group activities. We demonstrate that our approach significantly outperforms existing state-of-the-art techniques on the public group activity recognition datasets called collective activity dataset and augmented dataset. Visualized results (sample frames can be found in Appendix) can further demonstrate each input video frame with predicted bounding boxes on each human object and both predicted individual action and collective activities.