Efficiently Learning a Detection Cascade with Sparse Eigenvectors

TL;DR

This paper introduces a novel detection cascade training method combining boosting and sparse eigenvector techniques, achieving improved detection performance with computational efficiency.

Contribution

It presents BGSLDA, a new method that integrates boosting's re-weighting with GSLDA's class-separability for efficient cascade training.

Findings

GSLDA can be effectively used for object detection.

BGSLDA outperforms traditional boosting in detection accuracy.

The proposed method is computationally efficient.

Abstract

In this work, we first show that feature selection methods other than boosting can also be used for training an efficient object detector. In particular, we introduce Greedy Sparse Linear Discriminant Analysis (GSLDA) \cite{Moghaddam2007Fast} for its conceptual simplicity and computational efficiency; and slightly better detection performance is achieved compared with \cite{Viola2004Robust}. Moreover, we propose a new technique, termed Boosted Greedy Sparse Linear Discriminant Analysis (BGSLDA), to efficiently train a detection cascade. BGSLDA exploits the sample re-weighting property of boosting and the class-separability criterion of GSLDA.