How Accurate Is ARKit for Long-Distance Applications?
Recently while working on a request from a client, we were charged with figuring out how accurate ARKit is at longer distances. To test the framework, we created a simple app that lets you place a virtual pin object at a starting location and then begin measuring the distance between this point and a point on the ground that the device is oriented toward.
In order to test the app, we went out to a field that had markings every 15ft. for a total of 285ft. We placed our first pin and then went along the length of the field, marking the distance calculation made by the application.
For this test we used an iOS device with a LiDAR sensor. While we did not compare the results of this test with a device without the sensor, the actual application itself runs much faster, for example with horizontal plane detection.
Here’s a table showing the results of the test.
Notice the small amount of errors even among the shorter distances of 15–90ft. There is some amount of user error involved in taking these measurements. When using this app and orienting the device so that the cross hair intersects with the center of the field marking while you’re not directly above it, the angle of the device can make a large difference in the measurement. So there is apparently some user error involved in these results. As we go out even further than 90ft. it’s apparent that the increase in error between the measured distance and the actual distance comes from the drift of the sensors on the device and the loss of spatial awareness from ARKit. That being said, for these distances the results were impressive.
Note that this is probably the best case scenario for this setup as the ground was completely level and we traveled in a straight line to take measurements. However, if there were more physical ground features there’s a chance that ARKit would have an opportunity to better pinpoint its location relative to them.
Possible future testing could include:
- Non-linear paths
- Paths with significant slope
- Paths that included physical obstacles
Additionally, these same tests could be completed with a device without a LiDAR to judge the sensor’s effect on the results.
Hopefully these test results answer questions you may have regarding the accuracy of ARKit in long-distance AR applications.