Ranging Model and Algorithm Based on Monocular Vision for Autonomous Driving

Abstract

Aiming at the ranging problem in the autonomous driving scene, we present an algorithm model that employs the deep learning method to extract the vehicle body and its rear information. Then we introduce a monocular vision ranging algorithm that uses the camera projection geometry. In order to eliminate the impact of the road bumps, for detected targets, the width-distance mapping method is used to calculate the distance of targets. Comparing with camera geometric deduction ranging result, the pitch angle of the camera is computed, which is used for updating the camera extrinsic parameter. The test results show that our model can effectively measure the distance of the vehicle in front and the measurement error is no more than 10%, and the accuracy of the measurement results meets the requirements of early warning of traffic collision.