Emotion Recognition in Indian Classical Dance Performances: An Object Detection based Approach

Indian classical dance forms present nine emotions in contrast to six emotions which are considered as basic emotions. This work focuses on recognizing emotions of Indian classical dancers with special focus on Bharatanatyam dancers The paper introduces an emotion recognition model based on YOLOv9 (You Only Look Once version 9) model, an advanced framework for real-time object detection. There has been learning with extensive data augmentation and a composite loss function to manage class imbalance and improve generalization. Evaluated on the test dataset, the model achieves an average accuracy of 86%. The YOLOv9 based model achieves an Average Precision on the Mean (mAP) of 0.86.