Intelligent Assessment For Autistic Children Based On Multi-pose Fusion

Autism spectrum disorder(ASD),is a class of neurodevelopmental disorders.In China,the number of people with autism is at least 10 million,and is increasing at an annual rate of200000.In clinical autism screening,clinicians evaluate children based on assessment scales by observing their behavior in specific experimental tasks.This method takes a long time to test and has a complex evaluation process,requiring professional trained clinicians.The evaluation results lack objective and quantitative evaluation indicators,depending on the professional experience of clinicians,which may lead to certain errors.With the development of artificial intelligence technology,more and more intelligent assessment methods have been proposed.The intelligent assessment methods aim to capture the behavior patterns of subjects using computer vision technology,such as using computer vision related technology to perceive the subjects’ gaze,face,and head pose,and then achieve an objective quantitative evaluation for the response to name behavior of autistic children,reflecting their social interaction ability.Although the existing methods have achieved good performance,they ignore the physical behavior information of autistic children.Therefore,this paper proposes an intelligent assessment method for the response to name behavior of autistic children,which extracts multiple pose information through human pose tracking and head pose estimation technology for fusion.The research content and innovation points of this article are as follows:(1)This article has built a multi-view information collection platform,including four RGB cameras,to capture the behavioral patterns of autistic children undergoing clinical diagnosis from multiple perspectives and modes.We propose a “response to name dateset” and use automatic speech recognition and computer vision technology to achieve an objective quantitative evaluation for the response to name behavior of autistic children.In order to evaluate the response to name behavior of autistic children,human pose is innovatively used as the basis for evaluation.For videos from the left front of children,we use the human pose tracking algorithm to obtain the position information of children’s joint points.By analyzing the behavior patterns of children with positive and negative responses,and then using the child’s shoulder rotation angle and duration to comprehensively judge the response to name behavior.The consistency between our evaluation method and clinical diagnosis reaches 83.3%.(2)In order to further improve the accuracy of intelligent evaluation results,we make full use of the collected multi-view video information.For videos from behind children,we select a head pose estimation algorithm based on deep learning through experimental comparison to extract children’s head pose features.This paper proposes an intelligent assessment method for the response to name behavior of autistic children based on a combination of human pose and head pose,which involves automatic speech recognition,human pose tracking,and head pose estimation techniques.We use the child’s shoulder rotation angle,head pose information,and duration to comprehensively judge the response to name behavior.The consistency between our evaluation method and clinical diagnosis reaches 96.7%,which has been proved that the method proposed in this paper can play a good auxiliary role in early autism screening,providing valuable evidence for clinicians to diagnose.