Automated Tire Footprint Segmentation

Abstract

Quantitative image-based analysis is a relatively new way to address challenges in automotive tribology. Its inclusion in tire-ground interaction research may provide innovative ideas for improvements in tire design and manufacturing processes. In this article we present a novel and robust technique for segmenting the area of contact between the tire and the ground. The segmentation is performed in an unsupervised fashion with Graph cuts. Then, superpixel adjacency is used to improve the boundaries. Finally, a rolling circle filter is applied to the segmentation to generate a mask that covers the area of contact. The procedure is carried out on a sequence of images captured in an automatic test machine. The estimated shape and total area of contact are built by averaging all the masks that have computed throughout the sequence. Since a ground-truth is not available, we also propose a comparative method to assess the performance of our proposal.