Dynamic Belief Fusion for Object Detection

TL;DR

This paper introduces Dynamic Belief Fusion, a novel method that combines multiple object detection scores by estimating their uncertainty and dynamically assigning probabilities, resulting in significantly improved detection accuracy.

Contribution

The paper presents a new fusion technique, DBF, that effectively integrates multiple detectors using uncertainty estimation and Dempster's rule, outperforming existing methods.

Findings

DBF achieves higher detection accuracy than conventional fusion methods.

Experiments on ARL and PASCAL VOC 07 datasets validate DBF's effectiveness.

The approach effectively manages uncertainty in detection scores.

Abstract

A novel approach for the fusion of detection scores from disparate object detection methods is proposed. In order to effectively integrate the outputs of multiple detectors, the level of ambiguity in each individual detection score (called "uncertainty") is estimated using the precision/recall relationship of the corresponding detector. The proposed fusion method, called Dynamic Belief Fusion (DBF), dynamically assigns basic probabilities to propositions (target, non-target, uncertain) based on confidence levels in the detection results of individual approaches. A joint basic probability assignment, containing information from all detectors, is determined using Dempster's combination rule, and is easily reduced to a single fused detection score. Experiments on ARL and PASCAL VOC 07 datasets demonstrate that the detection accuracy of DBF is considerably greater than conventional fusion approaches as well as state-of-the-art individual detectors.