Human-Action-Recognition-with-Keras

Action/emotion detection is a very vital tool in smart home applications where the smart cameras can detect human actions/emotions using deep learning-based approaches and then address the user’s needs. In this study, a deep neural network for detecting the human action of “Eating” was developed using video clips that can be applied to smart cameras at homes. The widely available datasets from HMDB-51, UCF-101 and kinetics were used to train, test, and validate the models and then applied to YouTube videos recorded at home settings to examine the performance of the model in smart home applications. A combination of 3D CNN architecture for action detection and cv2 human face recognition resulted in an overall best accuracy for YouTube videos.