复杂交通条件下弱环境约束区域无人驾驶车辆相对定位方法关键问题的研究

项目名称： 复杂交通条件下弱环境约束区域无人驾驶车辆相对定位方法关键问题的研究

项目编号： No.61304194

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 姜岩

作者单位： 北京理工大学

项目金额： 23万元

中文摘要： 车辆与局部环境之间的精确相对定位是无人驾驶路径跟踪控制的基础。由于成本和可靠性的限制，高精度组合导航系统难以满足城市无人驾驶的要求。城市弱环境约束区域缺乏车道线等能够直接用于相对定位的环境元素，使得这类场景中的无人驾驶成为实现全自主无人驾驶的瓶颈。度量地图匹配是在弱环境约束区域中实现相对定位的有效手段。本课题解决在复杂交通条件下进行度量地图匹配时存在的两个关键问题：（1）避免运动障碍造成地图创建和匹配定位失败；（2）使匹配定位运算满足车辆行驶的实时性要求。研究内容包括：（1）研究无模型的多目标运动障碍检测与跟踪算法，对目标进行准确的运动估计，避免物体在视野中轮廓变化造成的速度跳动，同时保证不同物体运行轨迹交叉时仍然能够实现可靠的跟踪；（2）融合多模态度量地图，保证在激光栅格地图定位失败时仍然具有定位能力；（3）研究基于GPS增强的地图匹配定位算法，使匹配运算满足车辆运行的实时性要求。

中文关键词： 智能车辆；混合地图；同步定位与地图创建；马尔可夫定位；视觉里程计定位

英文摘要： Localization relative to its local environment is a prerequisite to the control of a autonomous driving vehicle. Due to the system cost and reliability in the positioning output, highly accurate integrated navigation system is hardly to satisfy the requirement of the autonomous driving in urban scenarios. The loose environmental constrained areas are lack of environmental elements which is able to localize the vehicle to its environment. Metric map matching-based localization is effective method for local localization in such areas. However, map matching methods are faced with two challenges when are applied to dense traffic scenarios: (1) avoid the moving objects to make the map construction and matching fail; (2) make the map matching satisfy the real time computation requirements for a continuously driving vehicle. This project is targeting at these two challenges. Design the model-free multiple moving objects detection and tracking algorithms which estimate the accurate velocity of objects and is resistant to the change in the shape of objects in the view of detection. Besides, the algorithm guarantees the correct tracking when the trajectories of objects merge together. Integrate multiple modes of metric maps to allow reliable localization ability when single metric is unable to work in highly dense traffic

英文关键词： intelligent vehicle；hybrid map；simultaneous localization and mapping；Markov localization；visual odometry localization