Vehicle logo recognition based on overlapping enhanced patterns of oriented edge magnitudes

Abstract

Vehicle logo recognition (VLR) has attracted wide attention from the community of intelligent transportation systems (ITS) due to its important role. Although many methods have been proposed for VLR, it remains a challenging problem. In this paper, we present a novel method for VLR. Our method includes (1) observation of the local anisotropy of vehicle logo images; (2) adoption of the idea of patterns of oriented edge magnitudes (POEM) and an advanced version of POEM for vehicle logo feature description called overlapping enhanced POEM (OE-POEM); (3) implementation of whitened principal component analysis (WPCA) for feature dimension reduction followed by collaborative-representation-based classification (CRC) as a classifier to perform VLR. We also construct a new vehicle logo dataset (HFUT-VL), which is larger and more comprehensive than the existing vehicle logo datasets. Finally, we conduct experiments on HFUT-VL, and the results indicate state-of-the-art VLR performance.