Occlusion -Aware Local Color Histograms for Real -Time Monocular ۳D Object Pose Tracking

We propose an enhancement to region -based, real -time monocular ۳D object pose tracking by introducing an occlusion -aware updating mechanism for local color histograms. Our method builds on the Gauss -Newton optimization framework for silhouette alignment but refines how local foreground and background color histograms are maintained over time. In particular, it detects and mitigates occluding objects or abrupt lighting changes that would otherwise corrupt the histogram statistics. By rejecting or down -weighting color samples that conflict with the existing distribution, our approach preserves accurate appearance models, thus improving robustness against partial occlusions and environmental variability. Experiments on semi -synthetic and real -world datasets indicate that this strategy yields more stable pose estimates in challenging, cluttered settings —all while retaining the low computational overhead characteristic of region -based tracking.

کلیدواژه ها:

Occlusion -aware tracking ، Real-time pose estimation ، Monocular tracking ، Region -based optimization ، Visual SLAM