Crowd Behaviour Recognition using Enhanced Butterfly Optimization Algorithm based Recurrent Neural Network

The crowd emotion recognition is a motivating research area that helps the security personals by means of the public emotions to interpret the crowd activity in a region. Approximately several conventional techniques exploit the lowlevel visual features to comprehend the behaviors of a crowd which widen the gap between the high as well as the low-level features. The objective model is used to expand the automatic algorithm for emotion recognition; hence this work uses the Recurrent Neural Network (RNN). The Bhattacharya distance is used for effectual emotion recognition, which is necessary to choose video keyframes. The keyframes are subjected to the Space-Time Interest Points (STI) descriptor which extracts features that structure input vector to the classifier. RNN is trained by exploiting the enhanced Butterfly Optimization Algorithm (Enhanced-BOA). The developed classifier identifies the crowd emotions, like Escape, Angry, Happy, Fight, Running/Walking, Normal, as well as Violence. The experimentation of the developed technique revealed that developed technique obtained a maximum accuracy, sensitivity as well as specificity, correspondingly.