Abstract

We propose a novel approach to increase the robustness of ob ject detection algorithms in surveillance scenarios. The cascaded con- fidence filter successively incorporates constraints on the size of the ob- jects, on the preponderance of the background and on the smoothness of tra jectories. In fact, the continuous detection confidence scores are analyzed locally to adapt the generic detector to the specific scene. The approach does not learn specific ob ject models, reason about complete tra jectories or scene structure, nor use multiple cameras. Therefore, it can serve as preprocessing step to robustify many tracking-by-detection algorithms. Our real-world experiments show significant improvements, especially in the case of partial occlusions, changing backgrounds, and similar distractors.