Human fall detection constitutes a crucial field of research, finding applications in healthcare, safety monitoring, and assisted living. Accurate and timely identification of falls can profoundly enhance the well-being and safety of individuals, particularly those who are vulnerable, such as the elderly or people with mobility impairments. Traditional fall detection methods often rely on supervised learning algorithms, which require substantial amounts of labelled data. However, acquiring high-quality labelled data for signals and time series, such as accelerometer readings, poses challenges due to the scarcity and cost associated with manual annotation.
In this study, we propose an innovative approach to human fall detection using inertial sensing and self-supervised learning. A SimCLR model is implemented from scratch and integrated with a simple neural network to form a binary classifier that can effectively distinguish between fall and non-fall instances even in a low data regime.