A Parallel Signal Detector Approach for Detection of Human Activities Using Multiple Seismic Sensors

Intrusion detection systems play an important role in the protection of vital installations and automated monitoring of unmanned areas. Seismic sensors such as geophones are used for sensing the ground vibrations caused due to human activities. These sensors are being employed for the covert detection of seismic signatures. This study investigates an in-depth analysis of different parameters of a signal detection algorithm STA/LTA for efficient detection of ground vibrations caused due human activities. High false alarm rates frequently limit the effectiveness of unattended ground sensor (UGS) systems, presumably as a result of flaws in the underlying algorithms and constraints in onboard computation. A real-time energy-based parallel recursive STA/LTA algorithm using a 32-bit analog-to-digital converter (ADC) with an ARM microprocessor is developed. The real-time detected event data is utilized for the optimization of the STA window length, and threshold values. Machine learning models such as support vector machines (SVM) with different kernels, k-nearest neighbor (kNN), and random forest are trained to classify the human activities. A STA window length of 300 samples with triggering thresholds (Ton = 2.7 and Toff = 1.8) achieves highest detection F1 score of ~ 0.74 with 81.7% accuracy of random forest classifier.