How can AI reduce fall injuries in the workplace?

TL;DR

This paper introduces a novel AI-based method combining RNN and KAN for real-time fall detection and impact timing estimation, addressing data heterogeneity and processing constraints in workplace safety applications.

Contribution

It presents a new approach using RNN and KAN for improved fall detection and impact timing estimation, tested on the SisFall dataset.

Findings

Achieved 82.6% true positive rate for falls

Achieved 98.4% true negative rate for non-falls

Estimated impact time with approximately 160ms error

Abstract

Fall-caused injuries are common in all types of work environments, including offices. They are the main cause of absences longer than three days, especially for small and medium-sized businesses (SMEs). However, data, data amount, data heterogeneity, and stringent processing time constraints continue to pose challenges to real-time fall detection. This work proposes a new approach based on a recurrent neural network (RNN) for Fall Detection and a Kolmogorov-Arnold Network (KAN) to estimate the time of impact of the fall. The approach is tested on SisFall, a dataset consisting of 2706 Activities of Daily Living (ADLs) and 1798 falls recorded by three sensors. The results show that the proposed approach achieves an average TPR of 82.6% and TNR of 98.4% for fall sequences and 94.4% in ADL. Besides, the Root Mean Squared Error of the estimated time of impact is approximately 160ms.