Investigating the Accuracy of Google’s Visual Positioning System

Viktor Horvath, Charles Toth, and Arpad Barsi

Peer Reviewed

Abstract:	In modern geolocation practices, Global Navigation Satellite Systems (GNSS) have effectively addressed positioning challenges in open-sky environments. However, the recent rise in jamming and spoofing incidents, along with intrinsic limitations of GNSS, particularly in GNSS-challenged or denied environments, has highlighted the need for alternative positioning solutions. Under such conditions, GNSS accuracy can degrade significantly, with real-time positioning errors ranging from a few meters to several tens of meters. Additionally, generating reliable ground truth in these environments is often difficult, which limits validation and development of alternative systems. To improve positioning accuracy in environments where GNSS signals are unreliable, landmark-based positioning methods offer a viable solution. These methods leverage the known coordinates of visible landmarks to improve location accuracy. This study evaluates Google’s Visual Positioning System (VPS), a cloud-based service designed to provide precise localization using smartphones. Accessible via the Google Cloud API (Application Programming Interface), VPS operates by uploading images captured by smartphones, along with approximate positional data, to Google's servers. The servers then match these images against its vast Google Street View database. Under favorable conditions, such as the presence of distinctive landmarks and high-quality imagery, VPS can achieve sub-meter level accuracy, surpassing traditional GNSS methods, even when augmented by additional sensors. Furthermore, VPS provides six-dimensional pose estimation, offering users enriched positioning data. In our first experiment, a small measurement track was established using control points measured with RTKGNSS as reference. A second experiment involved a longer route around a city block, where loop-closure and positioning accuracy were evaluated. Data collection was performed in both clockwise and counterclockwise directions, with positional data logged using a custom Android application. The recorded positions were subsequently analyzed and compared against the solutions provided by the Google VPS API and the reference ground control points. As an initial exploration of this topic, this study provides preliminary insights intended to stimulate further research into the potential of global landmark-based positioning systems. Keywords— camera localization, global positioning, smartphones, landmark-based positioning
Published in:	2025 IEEE/ION Position, Location and Navigation Symposium (PLANS) April 28 - 1, 2025 Salt Lake Marriott Downtown at City Creek Salt Lake City, UT
Pages:	1594 - 1600
Cite this article:	Horvath, Viktor, Toth, Charles, Barsi, Arpad, "Investigating the Accuracy of Google’s Visual Positioning System," 2025 IEEE/ION Position, Location and Navigation Symposium (PLANS), Salt Lake City, UT, April 2025, pp. 1594-1600.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In