Autism Spectrum Disorders (ASD), often referred to as autism, are neurological disorders characterised by deficits in cognitive skills, social and communicative behaviours. A common way of diagnosing ASD is by studying behavioural cues expressed by the children. We introduce a new publicly-available dataset of children videos exhibiting self-stimulatory (stimming) behaviours commonly used for autism diagnosis. These videos, posted by parents/caregivers in public domain websites, are collected and annotated for the stimming behaviours. These videos are extremely challenging for automatic behaviour analysis as they are recorded in uncontrolled natural settings. The dataset contains 75 videos with an average duration of 90 seconds per video, grouped under three categories of stimming behaviours: arm flapping, head banging and spinning. We also provide baseline results of tests conducted on this dataset using a standard bag of words approach for human action recognition. To the best of our knowledge, this is the first attempt in publicly making available a Self-Stimulatory Behaviour Dataset (SSBD) of children videos recorded in natural settings.

Self-Stimulatory Behaviours in the Wild for Autism Diagnosis Dataset

In recent years, it has been observed that many researchers have been working on different areas of detection, recognition and monitoring of human activities. The automatic determination of human physical activities is often referred to as human activity recognition (HAR). One of the most important technology that detects and tracks the activity of the human body is sensor-based HAR technology. In recent days, sensor-based HAR attracts attention in the field of computers due to its wide use in daily life and is a rapidly growing field of research. Activity recognition (AR) application is carried out by evaluating the signals obtained from various sensors placed in the human body. In this study, a new approach is proposed to extract features from sensor signals using HAR. The proposed approach is inspired by the Gray Level Co-Occurrence Matrix (GLCM) method, which is widely used in image processing, but it is applied to one-dimensional signals, unlike GLCM. Two datasets were used to test the proposed approach. The datasets were created from the signals obtained from the accelerometer, gyro and magnetometer sensors. Heralick features were obtained from co-occurrence matrix created after 1D-GLCM (One (1) Dimensional-Gray Level Co-Occurrence Matrix) was applied to the signals. HAR operation has been carried out for different scenarios using these features. Success rates of 96.66 and 93.88% were obtained for two datasets, respectively. It has been observed that the new approach proposed within the scope of the study provides high success rates for HAR applications. It is thought that the proposed approach can be used in the classification of different signals.

/pdf/a-new-approach-for-physical-human-activity-recognition-based-2im4y4x0mx.pdf

A new approach for physical human activity recognition based on co-occurrence matrices

Autism spectrum disorder (ASD) affects the cognitive development and communication skills of children and adults, limiting the functional capacity of these individuals. ASD has a worldwide prevalence of 1% on children and it affects not only the people with this disorder, but also their family and the surrounding community. In the family circle, individuals on the spectrum require greater support and attention relative to its cognitive capacity, impacting the mental and emotional health and even the financial life of families. The lack of infrastructure, trained professionals, and public health policies to deal with ASD is a known problem, specially in low income countries. To mitigate this issue, computer-aided ASD diagnosis and treatment represent a powerful ally, reducing the workload of professionals and allowing a better overall therapeutic experience. This paper intends to investigate how machine learning techniques can help specialists by providing an automated analysis of ASD recorded therapy sessions. The proposed solution is capable of handling large amounts of video data, filtering out irrelevant frames and keeping only relevant scenes for posterior analysis. Our results show that the proposed solution is capable of reducing manual checks by up to 51.4%, which represents a significant workload reduction for health experts. This solution will hopefully provide researchers, therapists and specialists with a tool that assists the automated identification of features and events of interest in video-recorded therapy sessions, reducing the amount of time spent in this task. 

Event Detection in Therapy Sessions for Children with Autism

Autism spectrum disorder (ASD) is a very high-flying area of research in the current era owing to its limited and on-going exploration. This chapter aims to bridge the gap of such late realization of autistic feature through machine intervention commonly known as computer vision. In this chapter, basic summarization of important characteristic features of autism and how those features could be measured and altered before a human could recognize are proposed. The chapter proposes a model for activity identification of the autistic child through video recordings. The approach is modelled in a way that consists of two phases: 1) Optical flow method detects the unusual frames based on motion pattern. 2) Each of these detected frames are fed to convolution neural network, which is trained to extract features and exactly classify if the particular frame under consideration belongs to usual or unusual class. This examines the various activities, time delay, and factors influencing the motion of the autistic child under constrained scenarios proving maximum accuracy and performance.

Activity Recognition System Through Deep Learning Analysis as an Early Biomarker of ASD Characteristics

Autism spectrum disorder (ASD) is one of the most notable neurodevelopmental disorders that gained major notification among parents, clinicians and even in researchers in the current era. The early identification of autism is a much needed support for parents and clinicians. The proposed methodology aims in building a computational model for such easy and early diagnosis by analyzing and finding the correlations between features-to-class and feature-to-feature so as to maximize the former and minimize the latter. The correlation between features is analyzed using (i) chi square computation technique in filter method and (ii) information gain. On analyzing the correlations, the resultant attributes of every technique are trained separately under the standard linear SVM classifier and then tested for the models performance and accuracy. There are two major contributions of the proposed work; Method 1: to build a model that takes optimized features extracted from the chi square and information gain analysis from questionnaires on the application of genetic algorithm (GA). The optimized features are then trained and tested to classify autism in support of SVM linear classifier. Method 2: to build a model based on the application of back-propagation feed forward neural network to classify the presence of autism. The paper ensures better and faster convergence of the positive class label of autism with maximized accuracy, specificity, performance and minimized error. The novelty of the paper lies in the fact of extracting important features for modeling so as to make a prior analysis by any parents at home before approaching clinicians which supports the early intervention of autism.

S. P. Abirami

Papers

Activity Recognition System Through Deep Learning Analysis as an Early Biomarker of ASD Characteristics

A Meta-Heuristic Model Based Computational Intelligence in Exploration and Classification of Autism in Children