Understanding Children's Gaze Behaviors
MetadataShow full item record
Identifying early signs of autism has been a challenging problem in the medical field. Many research studies aim to detect behavioral patterns of children with autism in the first three years of life. Early detection of autism allows early intervention to be initiated and thus is essential to achieving the best long-term outcomes for children with autism. Under the guidance of Drs. Jim Rehg and Agata Rozga, and mentored by PhD student Eunji Chong, I explored the challenge of developing automated measures to analyze children’s gaze behaviors during social interactions. We focused on using computer vision and deep learning techniques to analyze children’s attentions to objects and their social partner during play. In this thesis, I will first talk about related research works in the area of detecting autistic behaviors in children. I then introduce the two major pathways that I have taken to explore this problem, which are children’s head pose analysis and children’s gaze analysis. Details of the algorithm and the results are also presented and discussed. Finally, I will also address some failure cases and propose potential future works.