Tracking with inter-visibility variables

• Published in: IET Seminar on Target Tracking and Data Fusion: Algorithms and Applications pp. 59 - 68
• Main Authors: Horridge, P, Maskell, S
• Format: Conference Proceeding
• Language: English
• Published: Stevenage IET 2008
• Subjects: Other topics in statistics; Sensor fusion; Signal processing and detection; estimation theory; inter-visibility variables; sensor network; off-the-shelf multisensor tracker; probability; target tracking; sensor fusion; hypothetical target position; fixed probability
• ISBN: 0863419100; 9780863419102
• DOI: 10.1049/ic:20080057

This paper is concerned with the problem of tracking targets where each sensor in the network's view of the target is prone to obscuration due to other objects in the environment. An off-the-shelf multi-sensor tracker that attempts to localize (i.e. triangulate) targets moving through such a sensor network (for example, a CCTV network) assumes a fixed probability of detection for each sensor. This assumption is not valid when obscuration becomes pronounced and results in a degradation of performance. The tracker typically underestimates the probability of a track being present which causes premature track termination and thereby track fragmentation. One approach to countering this effect is to assume that a map of the environment is available and can be used to determine the probability of detection for any given hypothetical target position and given sensor position. This approach necessitates access to such a map, which may not be readily available. To address this issue, this paper build on the recent development of algorithms for estimating the probability of targets' existence (for track management) to propose a model that considers random variables associated with the time-varying inter-visibility of targets from sensors. We compare this proposed approach, the use of known maps and the use of off-the-shelf multi-sensor tracking algorithms using simulated data. Recommendations include the development of a framework for feedback to and from a perceived map, built up over time. The technical challenges associated with the robust sequential estimation of an unknown and fixed (or slowly varying) map are highlighted.