An effective vehicle logo recognition method for road surveillance images

In this paper, an improved vehicle logo recognition method is proposed for detecting and recognizing logo in images captured by road surveillance cameras. Firstly, a feature-match based symmetry detection algorithm is adopted for detecting symmetric axis in the image and rotating the image. Then a learning-based License Plate Localization (LPL) approach and a coarse logo location algorithm by using prior knowledge are used to obtain a coarse logo region. Moreover, a new vehicle logo location method based on Gabor Filters is put forward in order to localize the exact region of logo. In addition, two features- Pyramid of Histograms of Orientation Gradients (PHOG) and Multi-scale Block Local Ternary Patterns (MB-LTP) are combined to recognize the vehicle logo. At last, experimental results on a large number of images validate the robustness and efficiency of the proposed algorithm.