Social Visual Behavior Autism Flashcards
Three-brach-track network
- Head detection: detects heads in video frames
- Head Tracking: Uses a linking algorithm to link detected heads into tracks for proccessing
- Three-brach-tracking:
Head-pose-branches (2 branches): processes sequences of head crops to infer gaze direction
Head-Map-brach: encodes relative head positions, distances and the presence of other individuals.
Study Overview
Mutual Gaze Score graph
The result of the mutual gaze detection in four frames.
Mutual Gaze/ Human Coded Ratio Plot
The scatterplot displaying th regression lines and confidence intervals of the two differnt pla groups
Mutual Gaze Ratio
The proportion of frames where mutual gaze is detected (0 no mutual gaze, 1 constant mutual gaze)
Hand coded ratio
Frequency of mutual gaze as hand coded by experts, serving as the ground truth
What is the problem/motivation?
To introduce an social visual analytics behavior approach for observing the mutual gaze performance of children with autism
How was data collected?
- Collected video data from autism therapy sessions (ASD Dataset). The ASD dataset was used as input for the mutual gaze detection framework.
Discuss the Mutual Gaze Detection Framework
The development of the mutual Gaze detection framework is based on a three branch track network, a deep learning technique used to retrieve mutual gaze features between all possible pairs of participants in the video