An efficient hardware implementation of HON4D feature extraction for real-time action recognition

Human activity recognition has been an important area of computer vision research. In this paper, we present real-time hardware implementation for action recognition with HON4D features, which outperform the methods relying on skeleton detectors. Our proposed circuit adopts sliding histogram, and several approximate techniques to reduce computation and speed up feature extraction. Furthermore, using sliding histogram allows continuous classification without video segmentation in advance.