Treffer: A Deep Neural Framework for Self-Injurious Behavior Detection in Autistic Children.

Video Action Detection (VAD) is becoming increasingly common, with distributed methods shifting towards edge computing for real-time processing. The limited diversity and size of the existing Self-Stimulatory Behaviours Dataset (SSBD) hinder the gen-eralizability of Self-Injurious Behaviors (SIB) detection models. Early detection of SIB is imperative for timely intervention and support for individuals with Autism Spectrum Disorder (ASD), emphasizing the need for accurate recognition systems. Addressing these challenges requires advancements in dataset diversity, model accuracy, and generalizability to diverse populations and environments. This paper proposes a framework for the detection of SIB in children. We use a hybrid approach combining CNN and LSTM to detect SIB effectively. To enhance the recognition capabilities, we add new actions to the SSBD dataset to increase its size. This addition enriches the dataset, enabling more comprehensive training of our recognition models. We then extract frames from videos and apply augmentation techniques to the frames. The ConvLSTM, EfficientNet, and Long-Short Term Recurrent Convolutional Networks (LRCN) models are used for SIB action detection. Among these, the LRCN model demonstrates superior performance, achieving an accuracy of 92.62%, surpassing ConvLSTM (80.33%) and EfficientNet (77.17%). The LRCN model achieved a Mean Squared Error (MSE) of 0.045, highlighting its reliability in minimizing prediction errors for action detection. This underscores the effectiveness of hybrid models for video action recognition, emphasizing the importance of early detection in supporting individuals with ASD. [ABSTRACT FROM AUTHOR]