Analyzing Emotions of Real Autism Spectrum Disorder Patients Through Micro Expressions of Asd Characters in Movies: A Dataset Construction and Application Study

All over the world, the prevalence of Autism Spectrum Disorder ( ASD ) is increasing year by year. Moreover, autism spectrum disorder can only rely on rehabilitation treatment and cannot be cured with drugs. Through collecting expressions data of characters from nine ASD movies, 1040 data files are made and there are 454 classes of micro expressions in total. Based on Cinemetrics, this paper uses six kinds of object detection algorithms, which are Faster-RCNN, Single Shot MultiBox Detector (SSD), RetinaNet, YOLOv3, YOLOv5 and YOLOv8, to comparative analysis micro expression prediction models and apply them to medical research. After training the micro expression dataset, the two-stage Faster-RCNN gets the best training results. Faster-RCNN achieves 0.89 precision and 89.52 percent MAP. It can precisely predict micro expression of the character in the image, we recommend this algorithm model.