面向GNSS拒止与伪有效解场景的车载GNSS/INS/OSM融合连续定位方法

Continuous Vehicle Localization under GNSS Denial and Pseudo-Valid Fixes: A GNSS/INS/OSM ESKF Fusion Approach

摘要: 智能网联汽车在长大隧道与城市遮挡区段运行时，全球导航卫星系统(GNSS)观测可能中断。消费级接收机在定位有效标志(FixValid)仍为真时可输出坐标冻结型伪有效定位解，使松组合滤波持续错误更新。针对前期交互式多模型自适应融合(IMM-AF)常用实现未纳入完整惯性机械编排、未利用接收机速度，且未分离参考中心线与开放地图路径等不足，文章建立基于误差状态卡尔曼滤波(ESKF)的车载GNSS/惯性导航系统(INS)/地图融合方法。惯性测量单元(IMU)按平面运动作机械编排；伪有效定位解检测联立平面步进、滤波-GNSS偏差与u-blox导出速度，经连续确认后剔除异常解并转入地图约束航位推算(map-DR)。检测阈值与采样间隔、分段掩码共用同一套量；实现中冻结步进宽于运动学定义，是误报来源。理想化参考实验沿惯性探测器(IE)真值降采样中心线推进(ESKF + Oracle map-DR)，实际地图实验沿开放街道地图(OSM)推进，二者不合并解读。SensNav城区–隧道序列上，ESKF + Oracle map-DR全轨迹平面均方根误差(RMSE)为689.12 m，可用段15.64 m，异常段2031.79 m；GNSS单点定位为813.30 m，IMM-AF + Map为615.31 m、24.28 m、1813.35 m。2031.79 m度量检出后沿程积分的残余误差，属长时间GNSS异常下的方法局限，不是检测成功后的可用精度。同一框架下OSM路径全轨迹为1158.51 m，校园开阔为1057.34 m，由误报触发错误匹配放大。城区检测器召回率94.7%、误报率67.1%、F1值0.266。

Abstract: When intelligent connected vehicles travel through long tunnels and urban canyons, Global Navigation Satellite System (GNSS) observations may be interrupted. Consumer-grade receivers can keep FixValid true while outputting frozen, pseudo-valid fixes that contaminate loosely coupled filters. Building on a prior Interacting Multiple Model-Adaptive Fusion (IMM-AF) baseline that lacked full inertial mechanization, receiver velocity, and separated reporting of reference-centerline versus open-map paths, this paper formulates a GNSS/Inertial Navigation System (INS)/map fusion scheme based on a seven-state Error-State Kalman Filter (ESKF). An Inertial Measurement Unit (IMU) drives planar mechanization; a detector jointly tests planar step, filter-GNSS discrepancy, and u-blox speed, then withholds GNSS updates and switches to map-constrained dead reckoning (map-DR). Thresholds share the sampling interval and the segment mask; the implemented freeze-step is looser than the kinematic definition and is the source of false positives. An idealized experiment advances along an Inertial Explorer (IE)-derived centerline (ESKF + Oracle map-DR); a real-map experiment uses OpenStreetMap (OSM). On the SensNav urban-tunnel sequence, ESKF + Oracle map-DR attains a planar Root Mean Square Error (RMSE) of 689.12 m (available 15.64 m; anomalous 2031.79 m), versus 813.30 m for GNSS single-point and 615.31/24.28/1813.35 m for IMM-AF + Map. The 2031.79 m value is residual along-track map-DR error after detection, a limitation under prolonged GNSS anomaly rather than usable accuracy. The OSM branch yields 1158.51 m on the urban sequence and 1057.34 m on campus, driven by a false-positive mismatch. Detector recall is 94.7% with a 67.1% false-positive rate (F1 = 0.266).