Tests at five locations found that dim conditions stretch anchor errors past nine meters.
When an augmented reality object lands in the wrong place, users notice immediately. A sign that floats three meters from its doorway, or a label that drifts down the street, breaks the illusion faster than a slow frame rate. Getting the anchor point right is the whole game in location based augmented reality.
There are two common ways to place that anchor. The first fuses GPS and inertial sensors, which is quick to build but rarely precise enough outdoors. The second reads the camera image instead, using Visual Simultaneous Localization and Mapping, or VSLAM. A team had already paired VSLAM with Google Street View imagery so that the system matches what the camera sees against street level reference photos. What nobody had measured carefully was how ambient light changes the result.
The team ran 100 trials across five points of interest around Okayama University in Japan, using Android devices and the ARCore engine. Trials were sorted into four lighting groups: daylight, day overcast, day rainy, and low light. Each group held 25 datasets, so the comparison rested on equal footing.
Light intensity and accuracy move together. The analysis found a significant correlation below the 0.05 threshold, and illuminance explained roughly 42 percent of the variation in distance error. In bright conditions the anchor error hovered under a meter. In the low light group, errors ran past nine meters, which is the difference between a useful overlay and a confusing one.
Using confidence interval estimation, the team put a number on the practical floor: at least 434 lux is needed for feasible and consistent accuracy. Rain added a second problem, because wet surfaces change the visual references the tracker relies on, which pushed horizontal tracking time and accuracy around.
The takeaway is not that vision based AR fails. It is that the approach has a measurable operating range, and developers can design around it with adaptive strategies instead of discovering the limits during a demo. The team points toward dynamic light adaptation algorithms as the next step.
Komang Candra Brata et al. (2024). Performance Investigations of VSLAM and Google Street View Integration in Outdoor Location-Based Augmented Reality under Various Lighting Conditions. Electronics.
