Vehicle Distance Estimation Based on Stereo Camera System with Implementation on a Real ADAS Board

Abstract

In Advanced Driver-Assistance Systems (ADAS) one of the main tasks is to determine the distance of the objects in the vehicle's surroundings. This paper presents a solution that determines the distance of a detected vehicle using a stereo image pair, which is implemented on a real ADAS embedded platform (ADAS board). To improve the performance of the solution, after disparity map calculation, the presented solution implements Fast Global Smoothing (FGS) filter with guidance over the disparity map, which smooths disparity map surfaces that represent the same object of the scene, while preserving its edges. This solution is tested in a real environment using the real ADAS board and specific automotive cameras. Distinct vehicles located at different distances from the cameras are used in tests. The results show the proposed solution can provide high-performance results when implemented on a real ADAS platform while processing 7 frames per second for videos with a resolution of 1280×360 pixels. Furthermore, it is possible to conclude that despite the somewhat slower execution of the solution using the FGS filter with guidance, it makes sense to use it because, with still acceptable execution frequency, it provides a much better distance estimation, which is extremely important for ADAS.