Filtered Channel Features for Pedestrian Detection

TL;DR

This paper introduces a unified framework for pedestrian detection using filtered channel features and demonstrates state-of-the-art results on Caltech and KITTI datasets by combining low-level features with optical flow.

Contribution

It proposes a unifying framework for pedestrian detection with filtered channel features and systematically explores different filter families for improved performance.

Findings

Achieved top performance on Caltech and KITTI datasets.

Using only HOG+LUV features, the method performs strongly.

Adding optical flow features further improves detection accuracy.

Abstract

This paper starts from the observation that multiple top performing pedestrian detectors can be modelled by using an intermediate layer filtering low-level features in combination with a boosted decision forest. Based on this observation we propose a unifying framework and experimentally explore different filter families. We report extensive results enabling a systematic analysis. Using filtered channel features we obtain top performance on the challenging Caltech and KITTI datasets, while using only HOG+LUV as low-level features. When adding optical flow features we further improve detection quality and report the best known results on the Caltech dataset, reaching 93% recall at 1 FPPI.