Outdoor location based augmented reality puts digital labels on real places. Point a phone at a building and a floating card explains what it is. The hard part is getting that card to sit on the right spot. Most Android apps lean on GPS and a compass, and both drift. Labels land metres away from the door they should mark.
A team from Okayama University and Universitas Brawijaya tested a way to close the gap. They added visual SLAM and cloud based matching against Google Street View on top of the usual sensor fusion. The prototype runs as a native Android app through ARCore.
Visual SLAM tracks features across the camera feed and maps the surroundings as the user walks. When the tracker loses its grip, during a quick turn or on featureless pavement, the app sends a frame to a cloud service. That service matches it against Street View panoramas and pulls the device pose back to a known coordinate. GPS and inertial data keep the result steady between matches.
The evaluation used 10 points of interest around the Universitas Brawijaya campus in Malang. Each was measured several times and compared with the true coordinate using the Haversine distance.
The method needs Street View coverage. Where Google has not driven, the matching step has nothing to work with and the app falls back to sensor fusion alone. The extra accuracy is not free either: more threads stay busy and more memory is held, which matters on budget devices.
Next comes testing under different lighting conditions, and folding the approach into a full outdoor navigation build rather than a point of interest prototype.
Brata et al. (2024). An enhancement of outdoor location-based augmented reality anchor precision through VSLAM and Google Street View. Sensors, 24(4), 1161.
