Target Tracking with Non-standard Measurements

Abstract

Multi-target tracking is the process of estimating the number of targets and their states based on imperfect sensor measurements contaminated by noises, missed detections and false alarms. The key challenges lie in the uncertainty of the number of targets and the uncertainty of the correlation between the measurement set and targets. The multi-target tracking is expected to filter out the influence of these three aspects in order to obtain an accurate estimation of true target states. A Bayesian solution to this problem requires a model describing the association between measurements and the underlying (unknown) target states. The vast majority of traditional trackers (such as the JPDA or MHT tracker) use the so-called standard measurement model: each target generates at most one measurement and each measurement originates from at most one target at a given time, which is the well-known point target model. In fact, in addition to traditional trackers, the RFS-based multi-target tracking algorithms introduced in the previous chapters are also based on the standard measurement model. This model assumption greatly simplifies the development of multi-target trackers and is also consistent with most real-world situations. However, in some special cases, it is an unrealistic representation of the real measurement process, typically manifested in extended target tracking and target tracking with merged measurement.