Automatically predicting child engagement in dyadic interactions

1 in 68 American 8-year-old children are diagnosed with Autism Spectrum Disorders (ASDs). Though prevalent, ASDs are not typically diagnosed until children are older than 4 years. The results of clinical interventions are improved when children are diagnosed as young as possible but it is difficult to provide clinical evaluation and intervention for all children at such a young age. An automatic method that screened and flagged at-risk children could reach more children and would facilitate the speed of diagnosis. Engagement characterizes an individual's attention to and interaction with the people and objects in their environment. It is used by psychologists to measure a child's social development and a lack of engagement can signal delays, such as ASDs. We demonstrate the first methods capable of predicting the engagement of a child automatically. We apply computer vision techniques to predict child engagement during unscripted two-person interactions. We show that predicting engagement is a challenging task for automatic methods and non-expert people. The work in this thesis provides the first steps to creating an automatic screener for developmental delays.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2017

Author dc:creator

• Tsatsoulis, Penelope Daphne

Contributors dc:contributor

• Forsyth, David
• Hoiem, Derek
• Lazebnik, Lana
• Karahalios, Karrie
• Rehg, James

• Artificial intelligence
• Computer vision
• Attention
• Recognition
• Engagement
• Autism Spectrum Disorders

Statement dc:rights

• Copyright 2017 Penelope Tsatsoulis

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/97347

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/97347