A Deep Learning-Based Human Activity Recognition in Darkness

This work proposes a thermal camera-based human activity recognition approach using robust features and a deep recurrent neural network. To extract body skeletons from thermal images an opensource library is used, which is called OpenPose. OpenPose is a deep convolution neural network-based approach for multi-person key point detection. It is typically used on color images, but it is tried in this work on thermal images to see how it works in dark environment. After computing skeleton of the body, robust spatiotemporal features are extracted. Then, the robust features from the thermal videos are applied to a deep recurrent neural network for activity modeling and recognition. The proposed approach is very useful to monitor humans in dark environments where the other typical RGB cameras even fail to generate visually understandable images.