Mobile-GVIO: A Multi-Sensor GNSS-Visual-Inertial Dataset for Complex Urban Environments

Mobile-GVIO dataset is a fine-grained multi-sensor dataset for GNSS-Visual-Inertial fusion evaluation in complex urban environments. It provides synchronized monocular camera images, high-frequency IMU measurements, consumer-grade GNSS observations, and ground truth trajectories for seven real-world sequences across three environment types: indoor-outdoor hybrid, pure outdoor, and pure indoor. Data were collected using an Honor smartphone (visual-inertial, 1280x720 camera @30 Hz + 100 Hz IMU), an iPhone 11 Pro Max (WGS-84 GNSS @1 Hz), and a handheld LiDAR-IMU mapping system (ground truth via Fast-LIO2). The Honor and LiDAR-IMU are rigidly mounted on a stable frame, while the iPhone is carried alongside by the operator. Seven sequences:- IO-1: Indoor-outdoor hybrid, ~1005 m, 836 s- IO-2: Indoor-outdoor hybrid, ~616 m, 571 s- IO-3: Indoor-outdoor hybrid, ~720 m, 570 s- Outdoor-1: Pure outdoor, ~501 m, 394 s- Outdoor-2: Pure outdoor, ~889 m, 680 s- Indoor-1: Pure indoor, ~137 m, 149 s- Indoor-2: Pure indoor, ~91 m, 91 s Each sequence contains a ROS bag (.bag) with three topics: /cam0/image_raw (sensor_msgs/Image, ~30 Hz), /imu0 (sensor_msgs/Imu, ~100 Hz), /gnss0 (sensor_msgs/NavSatFix, ~1 Hz). Ground truth trajectories are provided in TUM RGB-D format, generated offline by Fast-LIO2. Calibration files (camera intrinsics, camera-IMU extrinsics, IMU noise parameters) for ORB-SLAM3 and VINS-Fusion are included. Indoor sequences have no valid GNSS data and serve as VIO-only baselines. Indoor-outdoor sequences feature corridors, squares, narrow passages, stationary periods, and environment transitions. Pure outdoor sequences include building blockages and GNSS multipath. For documentation and updates, see: https://github.com/SZU-Rob-IPNP-Lab/Mobile-GVIO-Dataset