Multi-Sensor Smartphone Mapping and Positioning: App, Dataset, and Benchmark

Smartphones have evolved into ubiquitous platforms equipped with a sophisticated array of sensors, including cameras, IMUs, GNSS receivers, LiDAR, and magnetometers. These sensors enable a wide range of applications, from navigation and mapping to virtual reality and accessibility tools for the visually impaired. However, despite this hardware potential, the development of robust multi-sensor fusion algorithms remains constrained by data limitations.

While existing public datasets—such as those from the Google Smartphone Decimeter Challenge (GSDC)—have advanced high-accuracy positioning by integrating GNSS and IMU data, they often overlook complementary sensors like cameras and magnetometers. Furthermore, current data collection tools (e.g., GNSSLogger, Sensor Logger) often lack the capability to log raw GNSS observations and visual data simultaneously with precise time synchronization. This gap hinders the application of emerging machine learning techniques that require diverse, synchronized input streams.

In this contribution, we introduce a custom Android application capable of collecting time-synchronized data from multiple sensors, including IMU, camera, GNSS, and magnetometer. We evaluate the time-synchronization capabilities of popular smartphone models, including the Google Pixel series, Xiaomi, Samsung, and OnePlus. Using this application, we compiled a comprehensive dataset in Calgary, Alberta, Canada, capturing diverse environments such as urban canyons, highways, parks, and farmland under varying weather conditions. The data includes both vehicle-mounted and handheld kinematic scenarios. Finally, to demonstrate the utility of the dataset, we establish a performance benchmark using conventional open-source software, such as VINS. This work provides the research community with a holistic benchmark dataset to advance multi-sensor fusion algorithms for smartphones.