Objectives: Our long-term aim is to develop technology interventions that help people on the autism spectrum to read facial-head gestures in real-time natural social interactions. To evaluate the efficacy of these technologies, an accurate baseline of recognition ability is needed, including testing dynamic expression recognition in familiar people and strangers.
Methods: We set up dyadic conversations including autistic-typical and typical-typical pairs, where each participant wore a tiny video camera to record facial and head movements. We collected over 100 videos of these natural conversations, and then segmented them based on expressive content. Ten typically developing adults, who scored well on the Eyes Test, independently labeled the expressions in the videos and rated them for difficulty. We then formulated a test using video clips with the highest inter-rater reliability. The videos were balanced for difficulty and for six generalization conditions (self, peer, familiar adult, stranger, unknown actor from Mind Reading DVD).
Results: We achieved high inter-rater reliability in the expressive categories of smiling, happy, thinking, unsure, and interested. Samples from these categories were used in a preliminary assessment of 8 typical people and 5 individuals aged 18-20 with a diagnosis of high-functioning autism. Scores varied greatly among the participants on the spectrum, some of whom scored just above random while some approached the abilities of well-scoring neurotypicals. Self vs. other, familiar vs. stranger and other results will be presented.
Conclusions: Our technology and method for evaluating expression recognition accuracy on self, peers, and strangers in situ could have several applications to future social skills training in persons on the autism spectrum.