Reliable classification of vehicle logos by an improved local-mean based classifier

Classification of vehicle logo is an important step towards the vehicle recognition that is required in many applications in intelligent transportation systems and automatic surveillance. A fast and reliable vehicle logo classification approach is proposed by first accurate logo detection, followed by an improved local-mean based classification algorithm. The recently published integrative logo detection method features of two pre-logo detection steps, i.e., vehicle region detection and a small RoI segmentation, which could rapidly focalize a small logo target. A two-stage cascade classifier proceeds with the segmented RoI, using a hybrid of Gentle Adaboost and Support Vector Machine (SVM), to generate precise logo positions. To address the issue of classification confidence which also facilitates a rejection option, we proposed an improvement on the local-mean-based nonparametric classifier and With a simple class posterior estimation, a rejection strategy becomes straighforward. A database of 15 different types of vehicle logos was created from images captured by surveillance cameras. The proposed scheme offers a performance accuracy of over 95% with a rejection rate of 8%, thus exhibits promising potentials for implementations into real-world applications.