Traffic Sign Recognition Using Affine Scale-invariant Feature Transform

摘要

This paper introduces a traffic sign recognition method using a new type of local image features. The Affine SIFT (Scale-invariant Feature Transform) feature are more affine invariant than SIFT in image scaling, translation, and rotation, tilt when camera view angles is variant. Firstly Keypoint are obtained from the input image using the ASIFT algorithm. Then feature parameters can be normalized as SIFT descriptors. Secondly, a Euclidean distance based nearest neighbor approach are employ to find candidate match in test image from the descriptors database obtained from the template images. Finally recognition result is achieved by voting based on low least-squares solution for candidate matched points. Experimental result shows that the traffic signs are correctly identified under different view, different scale, even in different illumination conditions.