Yizhou Wang and Lingyu Zhang
This project explores the traditional as well as novel approaches solving action detection problems. It is common to use neural networks which always cost a lot of time for training and testing. To solve this bottleneck of action detection, “Dict-Deep” and “Faster-C3D” architectures are proposed. Dict-Deep architecture adds feature extraction and over-complete dictionary learning steps before neural network. Then, Dict-Deep algorithm is implemented and tested on Weizmann and KTH human action dataset, which obtained 99.2% on Weizmann and 80.4% on KTH.