RFS Tracking Apparatus Reducing Hypothesis Combinations

The patent segments the tracking problem into two distinct phases: a prediction phase that uses motion models to forecast target states without considering observation points, and an update phase that selectively incorporates observation data. This segmentation allows the system to avoid the exponential complexity of traditional RFS tracking while maintaining accuracy by only processing relevant observations after prediction.

The patent performs prediction as a preliminary action before updating with observation data. By predicting target states in advance using motion models, the system establishes expected positions and velocities before comparing them with actual observations. This preliminary prediction reduces the computational burden during the update phase, as the system only needs to evaluate observations against predicted states rather than all possible combinations.

The patent applies local quality by treating prediction and update operations differently. The prediction phase uses motion models appropriate for each target's local characteristics (e.g., constant velocity for moving targets, static model for stationary objects), while the update phase selectively applies observation data based on local relevance. This localized approach reduces the need to generate hypotheses for all observation-point-tracker combinations.

The patent inverts the traditional tracking approach by performing prediction before updating, rather than updating with all observations first and then filtering. This inversion allows the system to predict target states independently of observation points, then selectively update only with relevant observations, thereby avoiding the exponential generation of hypotheses that occurs when all observation points are processed for each tracker simultaneously.

By performing prediction as a preliminary action, the system establishes expected target states before comparing with observations. This prediction step provides a reference framework that guides the subsequent update process, allowing the system to efficiently determine which observation points are relevant without having to evaluate all possible combinations, thus maintaining both accuracy and efficiency.

The patent implements feedback by using predicted states as the basis for selecting and weighting observation data during the update phase. The prediction provides expected values that serve as feedback for evaluating which observations are most relevant, allowing the system to accurately update tracker states while avoiding the computational burden of processing all observation points equally.