Kinesi: A Fall Detection Framework Converging 3 Axial Representation and UMAP Algorithm
Aditya Pratap, Devesh Sharma, Shanu Sharma · 2024
Individuals having motor-related disorders and elderly people often experience muscle weakness and decreased bone density, leading to an increased risk of falls. Accidental falls are a major cause of morbidity and disability in the elderly, with over one-third of them falling annually. Due to this, there is a significant demand for low-cost fall detection systems to reduce the risk of sudden impact of falls. In this paper, “Kinesi” an optimized fall detection framework is proposed to facilitate elderly or motor-disabled persons by detecting falls inside their home or in outside areas such as footpaths, roads, etc. The framework is based on data collected using a gyrosensor and accelerometer sensor, which is processed through a combination of 3 axial representations and a Uniform Manifold Approximation and Projection (UMAP) algorithm for optimized detection of falls. It can be incorporated into a wearable device to make targeted people self-dependent for their daily routine tasks. The proposed fall detection framework is implemented on Edge Impulse platform and the experimental results show that the algorithm achieves an accuracy of 92.5%.