Automatic Recognition of Affective Body Movement in a Video Game Scenario

This study aims at recognizing the affective states of players from non-acted, non-repeated body movements in the context of a video game scenario. A motion capture system was used to collect the movements of the participants while playing a Nintendo Wii tennis game. Then, a combination of body movement features along with a machine learning technique was used in order to automatically recognize emotional states from body movements. Our system was then tested for its ability to generalize to new participants and to new body motion data using a sub-sampling validation technique. To train and evaluate our system, online evaluation surveys were created using the body movements collected from the motion capture system and human observers were recruited to classify them into affective categories. The results showed that observer agreement levels are above chance level and the automatic recognition system achieved recognition rates comparable to the observers’ benchmark.