Global navigation satellite system real-time kinematic (GNSS-RTK) positioning is an indispensable source for providing absolute positioning for autonomous driving vehicles (ADV), due to its high accuracy when a fixed solution is achieved. Satisfactory accuracy can be obtained in open areas. However, the performance of GNSS-RTK can be significantly degraded by signal reflections from buildings, causing multipath effects and non-line-of-sight (NLOS) receptions. To fill this gap, this paper proposed a novel method to exclude the potential GNSS NLOS receptions, aided by the local environment description generated with 3D LiDAR and inertial sensor, to further improve the GNSS-RTK. The local environment description, the 3D point cloud map, is built via LiDAR/inertial integration using factor graph optimization. Then the potential GNSS NLOS receptions are detected and remove using the 3D point cloud maps before the GNSS-RTK positioning. Finally, the improved GNSS-RTK positioning is adopted to correct the drift of the 3D point cloud map derived from LiDAR/inertial integration. The effectiveness of the proposed method is verified through a challenging dataset collected in urban canyons of Hong Kong using the automobile-level low-cost GNSS receiver.