A Novel Localization Approach for Autonomous Vehicles Based on Map Matchingwith Curvature Features

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 reference 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 measurement 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 select 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 proposed 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.