面向智能网联汽车鲁棒连续定位的多源异构数据融合IMM-AF方法

An IMM-AF Method for Robust Continuous Localization of Intelligent Connected Vehicles Using Multi-Source Heterogeneous Data Fusion

摘要: 智能网联汽车的安全运行依赖高精度、高可靠性的连续定位能力。然而，含北斗的多星座全球导航卫星系统(GNSS)接收机在隧道及城市遮挡环境中仍可能输出“有效”但严重偏离的真值解，导致全轨迹误差统计被GNSS异常段主导。本文提出一种基于多源异构数据融合的IMM-AF + Map定位方法，融合GNSS、惯性测量单元(IMU)非完整约束、开放街道地图(OpenStreetMap, OSM)道路约束及隧道参考中心线地图航位推算，并采用异常解检测与剔除及分段精度评定方案。SensNav城区–隧道序列实验表明：GNSS可用段内GNSS单点定位均方根误差(RMSE)为7.60 m，全轨迹813.30 m；IMM-AF + Map全轨迹615.31 m (较813.30 m降低24.3%)，GNSS异常段1813.35 m (较2398.40 m降低24.4%)；异常解检测与地图航位推算为GNSS异常段恢复的关键(消融：关闭地图航位推算后全轨迹RMSE升至5073.72 m)。需指出：在GNSS可用段，OSM横向约束可能因道路几何偏差使IMM-AF + Map的RMSE升至24.28 m (高于无地图IMM-AF的7.45 m)，体现地图约束须与GNSS质量联动，RTK参考精度为0.35~0.73 m。

Abstract: The safe operation of intelligent connected vehicles depends on high-precision, high-reliability continuous positioning capabilities. However, GNSS receivers incorporating the BeiDou system may still output “valid” yet significantly deviated true-value solutions in tunnel or urban occlusion environments, leading to the overall trajectory error statistics being dominated by GNSS anomalous segments. This paper proposes an IMM-AF + Map positioning method based on multi-source heterogeneous data fusion, which integrates GNSS data, non-incomplete constraints from the Inertial Measurement Unit (IMU), road constraints from OpenStreetMap (OSM), and position estimation using the tunnel reference centerline map, along with an approach for detecting and eliminating anomalous solutions as well as performing segmented accuracy assessment. The SensNav urban-tunnel sequence experiment demonstrates that the Root Mean Square Error (RMSE) for GNSS single-point positioning within the available GNSS segment is 7.60 m, and the overall trajectory RMSE is 813.30 m; for the IMM-AF + Map method, the overall trajectory RMSE is 615.31 m (a reduction of 24.3% compared to 813.30 m), and the RMSE for the GNSS anomalous segment is 1813.35 m (a reduction of 24.4% compared to 2398.40 m). The detection of anomalous solutions and the application of map-based position estimation are critical for restoring the GNSS anomalous segment (ablation: when map-based position estimation is disabled, the overall trajectory RMSE increases to 5073.72 m). It should be noted that within the GNSS available range, the OSM lateral constraint may cause the RMSE of IMM-AF + Map to increase to 24.28 m (higher than the 7.45 m observed with the IMM-AF system without map data), demonstrating that map-based constraints must be integrated in conjunction with GNSS quality; the RTK reference accuracy ranges from 0.35 to 0.73 m.