高速动态背景下的夜间驾驶复合视觉感知方法及理论研究

项目名称： 高速动态背景下的夜间驾驶复合视觉感知方法及理论研究

项目编号： No.61273241

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： 明德烈

作者单位： 华中科技大学

项目金额： 80万元

中文摘要： 由于现代社会中汽车在人们生活中的不可或缺和夜间驾驶的危险性，"智能汽车"的研究和应用是近十年来各大公司和研究机构的热点问题，同时基于视觉导航的汽车夜视系统作为预防夜间事故的一个有效措施也成为汽车主动安全应用领域的研究热点。智能车辆达到真正的实用阶段所面临的主要瓶颈在于目标检测、辨识以及场景理解，尤其是夜间车载视觉感知问题更是待解决的难点。本项目拟采用红外/微光双模信息复合感知的视觉方法，充分注意图像序列的时变、动态和复杂场景等特点，将计算机视觉与机器学习相结合，建立目标模型、场景模型、推理规则，研究涉及到生物视觉、信息融合、人工智能等相关基础理论。具体研究内容与创新之处包括：(1)提出基于反馈机制的双模传感器夜视环境自适应感知方法；(2)提出基于目标建模与机器学习相结合的目标检测方法；(3)提出基于语义理解的道路环境认知及推理模型。研究成果将提升夜晚行车的安全性和舒适性。

中文关键词： 智能车辆；夜间驾驶；主动安全；计算机视觉；机器学习

英文摘要： Automobiles are playing a more and more important role in people's daily life of the modern society, but the danger of night driving is drawing much attention worldwide for lots accidents. Due to the contradiction of indispensability and hazard of automobiles, researches and applications of "intelligent automobile" has become a feasible technology which has been reseahched by companies and institutes in the past decade. As an effective method for preventing night-driving accidents, vehicle night viewing system based on visual navigation has been an imptortant parts of "intelligent automobile". Meanwhile,visual perception of night driving is a key problem to be solved. It is facing a main bottleneck which lies in target detection and recognition and scene interpretation. This project intends to adopt the visual methods of compound perception of dual-mode information (infrared and low-light-level) to conduct researches. In the process, time-various characteristics, dynamic features and scene complexity of image sequence will be paid adequate attention to confirm the validity of approaches. In addition, computer vision and machine learning will be combined to reconstruct target models, scene models and inference rules. Researches of this project refer to biological vision, information fusion, artificial intelligenc

英文关键词： Intelligent Vehicle；Night-driving；Active Safety；Computer Vision；Machine Learning