Handcrafted localized phase features for human action recognition

Human action recognition is one of the most important topics in computer vision. Monitoring elderly people and children, smart surveillance systems human-computer interaction are a few examples its applications. The aim this study to recognize human activities by utilizing phase information extracted from frequency domain video data as handcrafted features. Rather than estimating optical flow or computing motion vectors, we utilize localized descriptors dynamics scene. Phase correlation each two co-sited blocks consecutive frames clips were used train model using KNN classifier action. To evaluate performance our method, an extensive work has been done on three large complex datasets: UCF101, Kinetics-400 Kinetics-700. results show that approach succeeds recognizing actions across all these datasets with high accuracy.