Improved particle filter for multitarget-multisensor tracking with unresolved observations

Multi-sensor tracking of multiple targets, in clear but with some observations unresolved, is considered. The approach considered is based on the "particle filter", representing the target-state distributions by random samples. The improvements suggested here, namely choice of predicted samples and introduction of bounds on acceleration and speed, allow reduction of the number of mistaken detections and percentage of lost targets, as well as improving average tracking performance with unresolved observations.