Tire classification from still images and video

The use of different types of tires (e.g., all-season, snow, studded, summer) is regulated by law in several states and countries. Violation of tire usage laws typically results in substantial fines for infringers. In this paper, we propose an automated method to classify tires into snow, all-season and summer tires from still images or from a sequence of video frames. Our method first trains a Support Vector Machine (SVM) classifier on features extracted from a set of training images. Classification of test tire images is a two-stage process that entails feature extraction and tire classification based on the processing of the extracted features by the previously trained SVM classifier. The principle underlying the feature extraction stage is the representation of tire images via a low-dimensional approximation obtained from Principal Component Analysis (PCA). In order to improve robustness to changes in illumination and perspective, the features are extracted from the frequency representation of the binary edge map of the tire tread image. Our experimental results show that the proposed method achieves high classification accuracy.