Probabilistic matching of image- to model-features for real-time object tracking

Background clutter produces a difficult problem for edge matching within model-based object tracking approaches. The solution of matching all possible candidate image features with the model features is computationally infeasible for real-time tracking. The authors propose to draw probabilistic samples of candidate sets based on measures for local topological constraints. Line features have parallel and junction constraints. Continuous measures are used for evaluation of matching of the feature sets to avoid thresholds. This approach limits the number of matchings and processing time increases linearly with the number of features. Experiments show the correct selection among multiple candidates for different scenarios.