Publication

Abstract : Emotion recognition of special needs individuals is one of the challenges for those involved in the learning and care of such individuals. Misunderstanding of the true emotions of the special-needs individuals may lead to erroneous corrective actions which may further exacerbate their emotions/frustrations. This work aims to provide a solution to be deployed in learning environments with individuals with autism spectrum disorders (considered in this work) to aid the primary caregivers in understanding the emotional states for a group of special-needs individuals, such as those in a learning environment. The system is trained on the emotion data captured of the individuals in the group and then deployed to identify and display the emotions in a pictorial form to the primary caregiver on the attached display. Through this work, we were able to perform the emotion recognition for the selected group individuals with a high degree of accuracy (99.8%) without requiring large datasets (total of 125 training images). Additionally, the learned model was small and robust enough to be deployed on resource constrained platform ie. The raspberry pi, without the need for any cloud-based support. This ensures that the system works completely offline and thus allays any fears of intrusion of privacy.