UREDT: Unsupervised Learning Based Real-Time Footfall Event Detection Technique in Seismic Signal

Footstep-based systems have myriad uses like activity monitoring, person identification, and fall detection, to name a few. These highly perceptive systems employ seismic signals generated from human footfall events. Extraction of such events from a time-varying signal is a vital aspect of signal analysis. The essential quality of such an event detection technique lies in its capacity to retain its optimum performance in the face of ever-changing background noise in the signal. All the common off-the-shelf event detection techniques need to be adjusted to this environmental noise. Our proposed technique alleviates this problem as it requires no such repeated adjustment of parameters with varying environmental conditions. This technique has an inherent dynamic capability to adapt to fluctuating noise levels. We have compared our technique with amplitude, STA/LTA, and Kurtosis-based techniques. We have used the proposed technique to detect human footfall signals recorded by using seismic sensors. The performance of our technique has been tested on both indoor and outdoor seismic signals to study its behavior in the presence of various environmental noises. It has been found that our proposed technique outperforms all other aforementioned techniques in both scenarios.