AI-based Emotion Therapy Bot for Children with Autism Spectrum Disorder (ASD)

Abstract

Nonverbal cues that are sent through facial expressions are crucial in interpersonal communication. An individual's emotional state can be ascertained using a facial expression recognition system. In this proposed work, the system compares an input image with a trained dataset that is stored in the database to determine the emotional state of an image. A complicated neurodevelopmental illness called autism spectrum disorder (ASD) impacts communication, language, and social skills. Early identification of ASD individuals, particularly in children, would be advantageous to build and strategize the proper therapeutic strategy at the precise time. A person with autism spectrum disorders can be recognized by looking at their face, making eye contact and other facial features. Utilizing the training datasets and the deep learning models, the proposed method will be trained, tested and evaluated. The outcome of the proposed framework will be more accurate on autism face recognition based on neural networks and implementing the robot for monitoring. The performance analysis of three transfer learning architectures such as VGG16, MobileNet and Resnet50 were integrated with the therapy bot. The VGG16 architecture outperformed the existing transfer learning models with an accuracy of 97.66% on real-time images.