Vehicle type recognition based on adaptive scaling window and masks

In recent years, more and more approaches were proposed for vehicle logo recognition. However, most of the approaches achieve high performance only when the images have high resolution and the number of vehicle type to be classified is few. In this paper, a novel algorithm is proposed to treat with various resolutions of vehicle images and recognize a large number of vehicle logos. This algorithm is based on adaptive scaling sliding window and template matching with screening masks that is applied to detect the most accurate size and feature position of the target logo. In order to solve the problem that complicated texture noise affects the accuracy of template matching seriously, different screening masks are utilized for different vehicles. The algorithm avoids the difficulties of features localization and can recognize a great number of vehicle logos. This algorithm is applied on a dataset comprised of 10000 vehicle images with 102 types of vehicle logos taken in different environment by different traffic cameras. Experiment results show an overall recognition ratio of 91.62%.