Emotion estimation in crowds: the interplay of motivations and expectations in individual emotions

Providing an estimation of the emotional states of individuals increases the insights on the state of a crowd beyond simple normal/abnormal situations or behaviour classification. Methods intended for identifying emotions in individuals are mainly based on facial and body expressions, or even physiological measurements which are not suited for crowded environments as the available information in crowds is usually limited to that provided by surveillance cameras where the face and body of pedestrians can often suffer from occlusion. This work proposes an approach for analysing walking behaviour and exploiting the interplay of motivations and expectations in the emotions of pedestrians. Real-world data is used to test the prediction of motivations and annotations on the emotional state of pedestrians are added to evaluate the proposed method's capability to estimate emotional states. The conducted experiments show significant improvements over previous methods for estimating motivations and consistent results to the estimation of emotions.