Generating 3D dataset of gait and full body movement of children with autism spectrum disorders collected by kinect v2 camera

Abstract

Until now, a three-dimensional dataset combines gait and body movement analysis of children with Autism Spectrum Disorders (ASD) in controlled environments has not been published. ASD mean disorders of neurodevelopment that last a lifetime which occurs in early childhood and usually associated with unusual movement and gait disturbances. Threedimensional gait features captured by Kinect v2 can assist clinicians in diagnosing, clinical decision-making, and treatment planning of ASD. In this paper, Kinect v2 uses to build a 3D-skeleton-based gait dataset, which includes joints positions, the corresponding skeleton movement video, joints trajectories video captured by Kinect v2, and color videos captured by Samsung Note 9 rear camera. Besides building dataset, this paper classifies children with ASD from normal children by proposed system comprises four main stages: 1) Augmentation the database by using seven transformations to solve the problem of the small number of ASD cases, 2) Extracting features that we think play an important role in classification, 3) Reducing data dimensions using Principal Component Analysis (PCA) and 4) Using Multilayer Perceptron (MLP) to classify data. Classification accuracy when using eleven features result from PCA and MLP is 95% with 0.7 seconds to build the model.