Accurate and globally referenced positioning is the key prerequisite for fully successful autonomous systems. Global Navigation Satellite System (GNSS) could provide global positioning service in open areas. Unfortunately, under denes urban areas, in which it is typical for autonomous systems to operate, the performance of GNSS suffers from non-line-of-sight (NLOS) receptions and multipath effects caused by GNSS signal reflection and blockage from surrounding buildings. Thus, it will lead to position measurements with terribly large errors and GNSS outage. To facilitate long-term stable and accurate navigation, this paper proposes a method of continuous GNSS Real-time Kinematic (RTK) positioning aided by he 3D light detection and ranging (LiDAR)/inertial odometry (LIO) with intelligent GNSS selection in urban canyons. The coarse-to-fine LIO keeps on generating a locally accurate motion estimation and a registered point cloud map. Inspired by the outstanding capability of the environmental description of 3D LiDAR, available GNSS-RTK under light urban environment is detected as the degree of urbanization is evaluated via the sky-mask elevation angle calculation, which will be further integrated with the relative motion estimation from LIO based on factor graph optimization. Therefore, the continuous and accurate positioning is guaranteed by the local estimation from low-drift LIO in highly urbanized areas and global correction once the “Available GNSS-RTK” is obtained. The experimental results in a typical urban canyon in Hong Kong shows that the performance of the proposed integration pipeline is significantly improved in comparison with that without GNSS-RTK selection.