摘要
提出了一种新的基于曲率特征的自主车辆地图匹配定位方法,该方法通过计算自主车辆行驶轨迹和参考轨迹的尺度不变曲率积分特征及其相关性进行匹配,可以有效地消除因航迹推算(DR)传感器标定参数偏差和航向角估计偏差而引起的错误匹配问题.文中首先采用扩展卡尔曼滤波器融合惯性测量单元输出、方向盘转角和4个ABS(防抱死刹车系统)传感器测量的轮速,估计自主车辆的位姿状态,并据此从数字地图中选择匹配的候选路段.然后利用本文提出的曲率空间特征地图匹配算法实现路段匹配,并根据曲率和航向角变化确定匹配点,最后将其作为无迹卡尔曼滤波器的观测值更新滤波器,从而实现高精度的位姿估计.现场道路实验结果表明,该法能够有效地实现地图匹配,降低自主车辆DR产生的累积误差,从而能够在GPS(全球定位系统)信号失效情况下实现长距离精确定位.
Using the curvature features, a novel map-matching based localization approach for autonomous vehicles is proposed. By computing the scale-invariant curvature integral and its correlation of autonomous vehicle＇s historical and ref- erence trajectories for matching, the proposed approach can effectively eliminate the mismatch problem caused by odometer calibration parameters bias and azimuth estimation errors in dead-reckoning （DR）. Firstly, we integrate the inertial measure- ment unit output, steering angles, and wheel speed measurements from four ABS （anti-lock braking system） sensors by using the extended Kalman filter in order to estimate the autonomous vehicle＇s position and orientation, which are then used to se- lect the candidate matching segments from digital maps. Then, a map matching algorithm based on spatial curvature features is proposed to accomplish segment matching, and matching points are determined according to the changes in curvature and yaw. Finally, these matching points are further utilized as the measurements of the unscented Kalman filter to update the filter and achieve high-precision estimation of pose. The experimental results in the real road condition show that the pro- posed approach is able to realize map matching effectively, reduce the accumulative error of autonomous vehicles in DR, and estimate the pose of autonomous vehicles accurately for long-range navigation even if the GPS （global positioning system） signal occasionally fails.
出处
《机器人》
EI
CSCD
北大核心
2012年第4期440-448,共9页
Robot
基金
国家自然科学基金资助项目(90820305
60775040)
国家863计划资助项目(2011AA041001)
关键词
自主车辆
地图匹配
尺度不变特征
无迹卡尔曼滤波
autonomous vehicle
map matching
scale-invariant feature
unscented Kalman filter
