Inter-Vehicle Distance Estimation Method Based on Monocular Vision Using 3D Detection

Most autonomous vehicles build their perception systems on expensive sensors, such as LIDAR, RADAR, and high-precision Global Positioning System (GPS). However, cameras can provide richer sensing at a considerably lower cost, this makes them a more appealing alternative. A driving assistance system (DAS) based on monocular vision has gradually become a research hotspot, and inter-vehicle distance estimation based on monocular vision is an important technology in DAS. There are still constrains in the existing methods for estimating the inter-vehicle distance based on monocular vision, such as low accuracy when distance is larger, unstable accuracy for different types vehicles, and significantly poor performance on distance estimation for severely occluded vehicles. To improve the accuracy and robustness of ranging results, this study proposes a monocular vision end-to-end inter-vehicle distance estimation method based on 3D detection. The actual area of the rare view of the vehicle and the corresponding projection area in the image are obtained by 3D detection method. An area-distance geometric model is then established on the basis of the camera projection principle to recover distance. Our method shows its potential in complex traffic scenarios by testing the test set data provided on the real-world computer vision benchmark, KITTI. The experimental results have superior performance than the existing published methods. Moreover, the accuracy of occluded vehicle ranging results can reach approximately 98%, while the accuracy deviation between vehicles with different visual angles is less than 2%.