USDOT Number Localization and Recognition From Vehicle Side-View NIR Images

Orhan Bulan, Safwan Wshah, Ramesh Palghat, Vladimir Kozitsky, Aaron Burry; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2015, pp. 91-96

Commercial motor vehicles are mandated to display a valid U.S. Department of Transportation (USDOT) identification number on the side of the vehicle. Automatic recognition of USDOT numbers is of interest to government agencies for the efficient enforcement and management of the commercial trucks. Near infrared (NIR) cameras installed on the side of the road, to capture an image of an incoming truck, can capture USDOT images without distracting the drivers. In this paper, we propose a computer vision based method for recognizing USDOT numbers using an NIR camera system directed at the side of the commercial vehicles. The developed method consists of two stages; first, we localize the USDOT tag in the captured image using the deformable part model (DPM). Next, we train a convolutional neural network (CNN) using street-view house number (SVHN) dataset and sweep the trained classifier across the localized region. Based on the calculated scores, we infer the digits and their locations using a probabilistic inference method based on Hidden Markov Models (HMM). The most likely digit sequence is determined by applying the Viterbi algorithm. A data set of 1549 images was collected on a public roadway and is used to perform the experiments.