Human Action Recognition in Video using Non-Hierarchical Approach

The activity recognition and understanding in video by computer is defined as gaining knowledge about what is going on in video by analyzing the frames of video. The analysis of such kind by computer can be used for human – computer interaction, video monitoring, Identification based on behaviors and patients monitoring system. The main challenges faced for activity recognition are scale variation, occlusion, pause variation, anthropometric variation, execution rate. There are different methods available for each step of recognizing human activity coping with different challenges. The complex activity recognition methods also rely on the performance of primitive action recognition. For recognizing primitive action, the non-hierarchical approach is used. This approach represents the video as Bag of Words (BoW) model. We identify the limitations of this model and propose good practices for enhancing performance of the model.