Multi‑Sensor Fusion Positioning for Vehicle Navigation: GPS/IMU/Map‑Matching Solution

Gaode's positioning business consists of cloud‑based and on‑device modules. Cloud positioning handles Wi‑Fi fingerprinting, A‑GPS, trajectory mining and clustering, while on‑device positioning provides real‑time location for smartphones and vehicle‑mounted terminals. With the rise of urban canyon scenarios (high‑rise buildings, overpasses), users demand higher accuracy.

Vehicle‑mounted positioning faces two main advantages: richer sensor suites can be installed on the vehicle, and tightly coupled sensors enable high‑precision algorithm design. However, urban canyons remain a pain point because GPS signals are blocked or heavily interfered, leading to no fix or poor accuracy – an inherent limitation of active (GPS) positioning.

To mitigate this, GPS + IMU multi‑sensor fusion is increasingly adopted. IMU, as a passive sensor, compensates for GPS shortcomings, and additional sensors such as odometers and cameras can further enrich the fusion.

Gaode leverages its map data as the core of positioning. By combining GPS, IMU, odometer, and visual sensors with map‑matching (MM), a GPS/IMU/MM fusion solution (software + hardware) is proposed.

Pain points for vehicle‑mounted applications

Yaw drift: position points drift in tunnels or high‑rise areas due to signal blockage.

Loss of fix: low‑accuracy estimation in signal‑dead zones (e.g., parking lots, tunnels).

Road‑capture errors: incorrect selection of primary/secondary roads on overpasses.

The root causes are poor GPS accuracy and low‑quality dead‑reckoning (DR) due to sensor errors. Map‑matching alone cannot resolve road‑capture errors because it only constrains position.

Related terminology

GPS (Global Positioning System) provides high‑accuracy absolute positioning but requires line‑of‑sight. IMU (Inertial Measurement Unit) offers continuous attitude and acceleration data without line‑of‑sight, but its errors grow over time. Map‑matching (MM) aligns raw position estimates with road network geometry.

Technical solution

Two common industry approaches are compared (Table 1): a software‑only GNSS + MM solution (partial yaw‑drift mitigation, cannot solve loss of fix or road‑capture) and a hardware‑centric GNSS + IMU solution (partial yaw‑drift mitigation, can solve loss of fix, cannot solve road‑capture). Gaode integrates both, forming a soft‑plus‑hard GNSS + MM + DR scheme.

Three technology stacks are evaluated (Table 2): Satellite positioning (GNSS): global, low‑cost, but vulnerable to blockage. Map‑matching (MM): provides positional constraints, improves accuracy but lacks independent positioning. Inertial navigation (IMU/DR): continuous output, no external dependency, but error accumulates.

The fusion architecture consists of three layers (Figure 3): Data‑Adaptive Layer – standardizes and synchronizes sensor inputs. Aided Navigation Layer – computes intermediate results (e.g., GPS quality assessment, device error compensation, DR). Navigation Layer – merges outputs into a single reliable navigation solution using a Kalman filter.

Key modules

5.1 Basic modules GPS quality assessment: evaluates position, speed, heading, and reliability, categorizing data into GOOD, DOUBT, BAD, ABNORMAL. Device compensation: uses GPS to calibrate speed sensor scale factors and IMU biases, improving DR performance in GPS‑denied environments. DR algorithm: dead‑reckoning based on AHRS (Attitude and Heading Reference System) for attitude estimation and odometer integration for position. Fusion algorithm: Kalman filter fuses GNSS, MM, and DR outputs, leveraging GPS quality metrics to set observation noise.

5.2 Special features Main/auxiliary road recognition: extracts driving behavior features and computes transition probabilities to disambiguate parallel roads. Elevated road (overpass) detection: combines slope estimation from MM with pitch angle to avoid speed‑induced errors. Parking‑lot detection: identifies entry/exit and multi‑level transitions using lack of GPS, low speed, and elevation changes.

Effectiveness

Field tests validate the solution from two perspectives: (1) improvement on the three user‑reported pain points, and (2) comparison with competitor products and Gaode’s smartphone positioning.

6.1 Drift mitigation – multi‑reference fusion and scene classification significantly reduce yaw drift in tunnels and overpasses (Figure 14).

6.2 Device calibration – dynamic gyro bias estimation reduces heading error by >50 % and position error by >75 % in parking‑lot scenarios (Figure 15).

6.3 Road‑capture accuracy – achieved >90 % correct main/auxiliary road identification, surpassing a rival’s 75 %.

6.4 Horizontal and vertical comparisons – with less than 10 % of competitor hardware cost, Gaode’s solution improves the proportion of errors below threshold by 1‑5 % and road‑capture accuracy by 15 %.

Overall, the multi‑sensor fusion and map‑matching approach markedly enhances positioning accuracy in urban canyon environments, though the fundamental limitation of GPS blockage remains an open challenge.