SMU Data Science Review


In this paper, we present a detection algorithm that accurately differentiates the event of a person falling from normal Activities of Daily Living (ADL). Our algorithm processes signals recorded from accelerometers and gyroscopes built into wearable activity monitoring devices such as smart watches that are worn on an individual’s wrist. Existing algorithms are accurate but imprecise, and rely too much on inconveniently-placed sensors. We propose a pipeline that improves precision without sacrificing accuracy and ease of use. We present the use of a combination of threshold-based and machine learning-based approaches to develop a refined fall-detection algorithm that builds upon previous research. Using various pre-processing techniques such as magnitude and acceleration change vectors, our model labels the varied activities into falls and ADLs using k-means clustering. Finally, we test the accuracy of these labels in a Support Vector Machine (SVM) binary classifier. Our hope, given the potential danger of injury resulting from a fall, is to create an accurate and precise fall detection algorithm that could be the precursor to an autonomous emergency alert system.

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License