DBF: Dynamic Belief Fusion for Combining Multiple Object Detectors

In this article, we propose a novel and highly practical score-level fusion approach called dynamic belief (DBF) that directly integrates inference scores of individual detections from multiple object detection methods. To effectively integrate the outputs detectors, level ambiguity in each score is estimated using confidence model built on precision-recall relationship corresponding detector. For detector output, DBF then calculates probabilities three hypotheses (target, non-target, intermediate state (target or non-target)) based conditioned prior which referred to as basic probability assignment. The distributions over all detectors are optimally fused via Dempster's combination rule. Experiments ARL, PASCAL VOC 07, 12 datasets show accuracy significantly higher than any baseline approaches well used for fusion.