Research On Recognition In Children Autism Based On Realistic Interactive Scenarios

Autism Spectrum Disorder(ASD)is a neurological and developmental disorder characterized by social difficulties,communication challenges,and repetitive and stereotyped behaviors.Currently,the prevalence of autism spectrum disorder is increasing worldwide,and the treatment situation remains critical.Early warning and diagnosis are important prerequisites for effective treatment on patients with autism spectrum disorders,as timely intervention can greatly improve the prognosis of patients.Due to the lack of a reliable diagnostic biomarker,the diagnosis of autism in the clinical setting relies mainly on interview-based diagnosis by professional psychiatrists using scales.This type of diagnosis is highly accurate and commonly applied,but has limitations such as time-consuming,strict implementation criteria,and strong subjectivity.People with autism spectrum disorders commonly exhibit atypical cognitive behaviors such as inflexibility,difficulty in shifting attention or changing behaviors,and restricted interests.With the rapid development of computer technology,computeraided diagnosis technology with a core of artificial intelligence is being widely used.Compared with traditional interview-based diagnosis,computer-aided diagnosis has a set of advantages such as being non-invasive,objective,natural and unconstrained.In order to solve the problems of the existing related studies with a single acquisition scenario and paradigm,this thesis proposes a standardized experimental paradigm based on realistic interactive scenarios to explore atypical patterns of cognitive behaviors such as social and perceptual behaviors in children with autism,and to achieve early detection of autism based on artificial intelligence algorithms.The specific research contents and innovations of this thesis are as follows:(1)Based on the cognitive characteristics of children with autism,the thesis designs a set of scenario-based spatial paradigms,including person/object preference,joint attention,and vocal response.These paradigms are executed in a three-dimensional space to simulate realistic interaction scenarios,then the atypical social and perceptual behavioral performance of children with autism will be record through video streaming for behavioral analysis and assessment of autism.This thesis also designs and develops a behaviors acquisition system for the automated control on the experimental process of the scenario-based spatial paradigm and the collection of children’s behavioral data during the paradigm execution.(2)To analyze the social and perceptual behaviors and explore the differences in the cognitive-behavioral patterns of children with autism,the head and eyes features of children during paradigm execution are extracted.This thesis proposes a 5-point head pose extraction method based on SQPn P to calculate the head pose of the subject children in space,which solves the problem that the traditional 64-point head pose estimation method fails to extract the head pose in the case of large angles or small faces.Then,based on the head pose,two normalized eye images are extracted from the face image,and the gaze angle of both eyes is predicted using a Res Net gaze estimation neural network.Finally,the atypical cognitive characteristics and behavioral patterns of children with autism are explored based on the changes of children’s head pose and eye gaze during paradigm execution.The results of the analysis show that children with autism exhibits a lack of joint attention,weak responses to social stimuli,and delayed responses to auditory stimuli in the scenario-based spatial paradigm,which are significantly different from typically developing children.(3)To achieve early warning and assessment of autism,this thesis proposes a multiscale behaviors fusion framework for the recognition of autism in children based on unimodal video data.Based on scale-space theory,this framework extracts multiscale behavioral features from head,eye,and limb,achieves information complementarity between different behavioral scales based on methods such as feature fusion and decision fusion,and uses multiple time-series models for validation.The results show that multiscale behavioral fusion effectively improves the accuracy of autism recognition in unimodal video data,with the highest accuracy of 95.26% and the highest F1-score of 95.23% for 10-fold cross-validation.This thesis also conducts a visual analysis of the feature distribution of multiscale behavioral features,and the analysis results show that in the high-dimensional feature space,the feature distribution characteristics of the autistic children group were significantly different from those of the typically developing children group.In summary,this thesis proposes a cognitive-behavioral assessment system for early and rapid recognition of children autism.The thesis designs a set of scenariobased spatial paradigms and develops a behaviors acquisition system as the basis for behavioral data collection.Then,the behavioral data of children in multiple paradigms are analyzed to explore the atypical cognitive behavior patterns of children with autism in the stimulation of the scenario-based spatial paradigm.Finally,the accuracy of autism recognition on unimodal video data is effectively improved by extracting multiscale behavioral features of body for fusion modeling.