面向无人驾驶汽车的视觉道路环境感知算法研究

项目名称： 面向无人驾驶汽车的视觉道路环境感知算法研究

项目编号： No.61272062

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 肖德贵

作者单位： 湖南大学

项目金额： 80万元

中文摘要： 本项目拟针对无人驾驶汽车道路环境感知的特定应用，从无人驾驶汽车的角度，研究结合无人驾驶汽车行车参数和计算机视觉的道路检测与道路行驶环境中障碍物的检测。提出一种结合行车参数确定视觉处理感兴区域的方法，以提高算法处理速度；提出一种结合行车参数、视觉消失点、道路表面特征及多帧视频图像的道路检测算法，以提高道路检测的精度；提出一种结合行车参数、道路检测结果和极坐标变换的障碍物检测算法，以实现障碍物的快速检测；提出一种结合行车参数的分级并行区域粒度自适应划分方法，实现算法的并行化，针对道路检测和障碍物检测的特点，分别采用矩形区域和扇形区域；设计一套完整的平台与算法的试验与验证方案，经过三级平台验证：通用PC及并行分布式平台、嵌入式计算及其并行分布式平台、无人驾驶汽车平台，实现算法的实时性、稳定性和有效性。针对无人驾驶汽车具体平台的视觉道路环境感知算法，预期将促进推广无人驾驶汽车的实际应用。

中文关键词： 无人驾驶汽车；视觉计算；目标检测与表达；高级驾驶员辅助系统；嵌入式计算

英文摘要： This project is planned for the specific application of road environment perception of UGV. Starting from the UGV's point of view, this project researches vision-based road detection algorithms and road obstructions detection algorithms by combining computer vision with driving parameters of UGV. To increase processing speed, a method is proposed to determine the region of interest in a image according to UGV's driving parameters. In order to increase road detection precision, a road detection algorithm, which integrates road surface features, vanishing point and multi-frames with UGV's driving parameters,is put forward. In order to detect road obstructions quickly, a road obstructions detection algorithm, which integrates driving parameters, road detection results and polar coordinate transform, is studied. To design parallel detection algorithms, a hierarchical and adaptive partition of parallel regions granularity is proposed. And aiming at different characteristics of road detection and obstructions detection, rectangle regions and fan regions are adopted respectively. In the project plan, a series of testing and verification programs for the specific platform and those algorithms are designed to improve the real-time, stability and validity of those algorithms through three levels verifyication and validati

英文关键词： unmanned ground vehicle；vision computing；object detection and representation；Advanced Driver Assistance Systems；embedded computing