Spectrum Analysis Techniques for Personnel Detection Using Seismic Sensors

摘要

There is a general need for improved detection range and false alarm performance for seismic sensors used for personnel detection. In this paper we describe a novel footstep detection algorithm which was developed and run on seismic footstep data collected at the Aberdeen Proving Ground in December 2000. The initial focus was an assessment of achievable detection range. The conventional approach to footstep detection is to detect transients corresponding to individual footfalls. We feel this is an error-prone approach. Because many real-world signals unrelated to human locomotion look like transients, transient-based footstep detection will inevitably either suffer from high false alarm rates or will be insensitive. Instead, we examined the use of spectrum analysis on envelope-detected seismic signals and have found the general method to be quite promising, not only for detection, but also for discrimination against other types of seismic sources. In particular, gait patterns and their corresponding signatures may help discriminate between human intruders and animals. In the APG data set, mean detection ranges of 64 meters (at P_D=50%) were observed for normal walking, significantly improving on ranges previously reported. For running, mean detection ranges of 84 meters were observed. However, stealthy walking (creeping) remains a considerable problem. Even at short ranges (10 meters), in some cases the detection rate was less than 50%. In future efforts, additional data sets for a range of geologic and environmental conditions should be acquired and analyzed. Improvements to the detection algorithms are possible, including estimation of direction of travel and the number of intruders.