Towards Automatic Engagement Recognition of Autistic Children in a Machine Learning Approach.

Zihan Tan,Anjia Zhou,Xiaojun Hei,Yayu Gao,Chengwei Zhang
DOI: https://doi.org/10.1109/tale48000.2019.9225935
2019-01-01
Abstract:Children suffering from autism have been continue rising globally and the early intervention may significantly improve the social and cognitive capacity of them. The autism treatment in China is still largely based on training courses and interventions, nevertheless, lacking of professional teachers is a severe problem. Traditional individual training classes are mainly based on teachers' evaluation of students' emotional feedback, and there remain prejudices. The continuous information technology advance, such as ML, have been inspiring novel diagnosis and treatment for autistic children. Note that the majority of the current research on emotional recognition of autistic children have not been conducted on the autistic children in China, in this paper, we aim to conduct a emotional recognition study for them. We study the one-to-one autism training class, and propose an automatic engagement recognition framework to track the behaviors and affection of autistic children by ML. With the improved spatial temporal graph convolutional neural network, our experiments based on the available datasets show that the behavior recognition accuracy reaches beyond 90% and the affection recognition achieves 99%. The result demonstrates that the proposed system can analyze the relationship among arousal, valence and expression category for the autism training class.
What problem does this paper attempt to address?