Proceedings Article | 22 October 2013
KEYWORDS: Sensors, Roads, Acoustics, Target detection, Signal processing, Monte Carlo methods, Sensor networks, Motion models, Detection and tracking algorithms, Signal detection
Tracking ground targets using low cost ground-based sensors is a challenging field because of the limited capabilities of such sensors. Among the several candidates, including seismic and magnetic sensors, the acoustic sensors based on microphone arrays have a potential of being useful: They can provide a direction to the sound source, they can have a relatively better range, and the sound characteristics can provide a basis for target classification. However, there are still many problems. One of them is the difficulty to resolve multiple sound sources, another is that they do not provide distance, a third is the presence of background noise from wind, sea, rain, distant air and land traffic, people, etc., and a fourth is that the same target can sound very differently depending on factors like terrain type, topography, speed, gear, distance, etc. Use of sophisticated signal processing and data fusion algorithms is the key for compensating (to an extend) the limited capabilities and mentioned problems of these sensors. It is hard, if not impossible, to evaluate the performance of such complex algorithms analytically. For an effective evaluation, before performing expensive field trials, well-designed laboratory experiments and computer simulations are necessary. Along this line, in this paper, we present an object-oriented modeling and simulation framework which can be used to generate simulated data for the data fusion algorithms for tracking multiple on-road targets in an unattended acoustic sensor network. Each sensor node in the network is a circular microphone array which produces the direction of arrival (DOA) (or bearing) measurements of the targets and sends this information to a fusion center. We present the models for road networks, targets (motion and acoustic power) and acoustic sensors in an object-oriented fashion where different and possibly time-varying sampling periods for each sensor node is possible. Moreover, the sensor’s signal processing and detection blocks are modeled using a parametric approach by associating a receiver operating characteristics (ROC) curve to the whole process, which results in false alarms as well as missed detections. The proposed simulation environment can be used for ground-truth and synthetic data generation for road-constraint multiple target tracking in an unattended acoustic sensor network.
