With the rapid advancement of AI and sensor technologies, autonomous driving is approaching commercialization. However, reliability remains a major barrier. Virtual-real fusion testing presents a promising solution, though existing methods often suffer from low modeling accuracy and reliance on GNSS. To address these challenges, we propose an autonomous driving testing system that integrates Augmented Reality (AR) and Visual-Inertial Simultaneous Localization and Mapping (VI-SLAM). By combining the ORB-SLAM3 framework with RGB-D dense mapping, our system enables high-precision environmental modeling without the need for GNSS. Additionally, we utilize the optical flow-based optimization to enhance virtural-real fusion. Experimental results demonstrate that our method can accurately align the virtual and real environments and can provide a reliable and high-fidelity testing framework for autonomous driving.