An Efficient Method for Vehicle Model Identification via Logo Recognition

A novel method is proposed for efficient vehicle model identification. Considering logo recognition vital, the method begins with the development of an algorithm based on image gradients for initial or coarse positioning of vehicle logo, then proceeds to blur detection, pixelization and Sobel algorithm for fine positioning of logo, and determines process via a SVM classifier based on DCT low-frequency characteristics. Local image on logo's left is chosen as ROI, its low-frequency characteristics are extracted and then fed to a SVM classifier. A dataset of 1096 front and side images were collected from 12 vehicle models entering a highway inspection station. Experiments show that average vehicle model recognition accuracy is 97% within 30ms recognition timespan.