Optimized Fall Detection Algorithm with Adaptive Sum Vector Magnitude and Axis-Weighted Features from Wearable Accelerometer Data

This study presents a fall detection approach utilizing body-worn accelerometer data. The proposed method initially involves segmenting accelerometer data into non-overlapping windows. Emphasis is placed on the most informative axis for fall recognition, determined by the axis exhibiting the most prominent features. This is achieved by calculating axis weights based on three crucial features: standard deviation, kurtosis, and entropy. The weight for each axis is computed using a weighted average formula. A novel metric named Adaptive Sum Vector Magnitude (ASVM) is introduced, derived from accelerometer data. This metric takes into account accelerometer readings along the x, y, and z axes. To assess the performance of the ASVM metric, a comparative analysis is conducted with two other metrics: Sum Vector Magnitude (SVM) and Differential Sum Vector Magnitude (DSVM). This work demonstrates the potential of the proposed ASVM metric for fall detection and provides insights into the prioritization of axis data through feature weighting.