Fast Monocular Vision Based Vehicle Distance Measurement by License Plate Localization

In this paper, we propose a fast monocular vision based vehicle distance measurement method based on the vehicle and license plate detection. This method mainly comprises of three modules: vehicle detection, license plate localization and distance measurement. Firstly, the vehicle detection module relies on the Faster-RCNN algorithm, with different scale and aspect ratios for the region proposal network (RPN) by the analysis of KITTI datasets. The experimental results on the KITTI dataset show that the modified Faster-RCNN method can improve the average precision (AP) and perform even better than feature pyramid network (FPN). Secondly, we localize the license plate from the detected vehicle with basic image segmentation strategies, which can accurately segment out the license plate. Finally, Due to the fixed size of the general license plate, we can make a measurement for the distance between the frontal vehicle based on the camera interior parameters and geometrical reasoning. Experiments on our unmanned vehicle platform show that the proposed method can calculate the distance accurately within the safe distance of 30 meters.