Emotion Recognition via Body Gesture
Son Thai Ly, Guee-Sang Lee, Soo-Hyung Kim, Hyung-Jeong Yang · 2018
Many psychological research revealed that bodily gestures convey crucial information to emotion recognition. This valuable aspect, however, has not gained much the attention from the science community, compared to other modalities, such as facial expression and speech in the emotion recognition task. There are limited works on this topic. Unlike recent works which exploited the hand-crafted features, this study proposes an end-to-end deep learning approach for gesture-based emotion recognition. Firstly, we adopt the hashing method to extract the keyframes from the video. Secondly, a convolutional LSTM network is used for exploiting the sequence information. Our results exceed most of the hand-crafted results, it achieves the state-of-the-art results for end-to-end deep learning based technique for gesture-based emotion recognition on FABO dataset. The research also indicates the promise for future improvement.