Study on Surround View System

Abstract

In this thesis, we propose some advanced features of a surround view system that displays the surroundings of the vehicle to help drivers drive more conveniently and safely. First, we propose a memory-efficient architecture for a surround view system. In order to provide surround views of higher resolution, a significantly high memory bandwidth is needed. For example, a surround view system with full high-definition (HD) resolution (1,920 × 1,080) requires six times the memory bandwidth required for a system of D1 resolution (1,280 × 720). In addition, a higher amount of single writes degrades the effective performance of a DRAM. To address these problems, two methods are proposed in this thesis. The first method reduces the required memory bandwidth by storing in the memory only the input pixel data that will be used in the final processing step in the memory, and the second improves the single-write performance of a DRAM subsystem by DRAM-aware data mapping. The effectiveness of the proposed methods is proved by designing a surround view system incorporating five field-programmable gate arrays and six DDR2-200 SDRAMs. In order to implement the proposed methods, we design new LUTs for pixel-level mapping. The proposed methods reduce the effective memory bandwidth requirement by 51%, allowing a full-HD surround view system to run at over 24 frames per second (fps). A bird’s-eye-view image synthesized by traditional surround view systems is projected on a ground plane. In this image, non-ground objects appear stretched toward the outward direction, and the observable area is relatively small. In addition, non-ground objects near the seam line disappear or are duplicated. We propose and formulate a three-dimensional (3D) mapping method that provides a more realistic view to drivers. This method uses semi-sphere or bowl shape as mapping surfaces instead of the ground plane. Then, we illustrate some examples of 3D surround view mapping. Finally, we propose two surround view calibration methods. One is called an in-line calibration that uses some simple and segmented patterns instead of the chessboard pattern of the existing calibration methods. This calibration requires relatively small space and it is robust to illumination change. Another method is called an on-road calibration that uses lane marking to adjust the change in the camera’s alignment due to the change in tire pressure or camera position. Both methods are implemented on an embedded system and tested on a real vehicle.