​

3D Vision Aided GNSS Real-time Kinematic Positioning for Autonomous Systems in Urban Canyons

​

​

Abstract:

The combination of the visual/inertial system (VINS) and global navigation satellite systems (GNSS) real-time kinematic (RTK), namely GNSS-RTK, is a promising solution to provide accurate, cost-effective, and drift-free positioning service for autonomous systems. However, its performance is significantly degraded in urban canyons due to the poor GNSS measurement quality and satellite geometry distributions, caused by signal blockage and reflections from surrounding buildings. In this paper, we propose a 3D vision-aided method to improve the GNSS-RTK positioning by detecting the potential reflected outlier GNSS signals and improve the satellite geometry distribution using low-lying visual landmarks. To detect the blocked but reflected GNSS non-line-of-sight (NLOS) receptions, a sky-pointing camera together with the deep neural network is employed to capture and segment the sky and non-sky view to therefore excluding the outlier GNSS measurements. However, this can cause the other new problem, the degraded satellite geometry distribution. To fill this gap, this paper explores the complementariness of low-lying visual landmarks and the healthy but high-elevation satellite measurements to improve the geometry constraint, where the IMU measurements, visual landmarks captured by a forward-looking camera, healthy GNSS measurements are tightly integrated via sliding window optimization to estimate the float solution of GNSS-RTK together with the associated covariance. Then the integer ambiguities and fixed GNSS-RTK solution are resolved based on the estimated float solution and covariance. The effectiveness of the proposed method is verified through several challenging datasets collected in urban canyons of Hong Kong using geodetic level and low-cost automobile level GNSS receivers together with automobile visual/inertial sensor suit.

 

​