Hi everyone,
Our team is currently using Cesium for Unreal to showcase built environment projects. To ensure the reliability of our visualizations and potential coordination tasks, we are looking for specific documentation or white papers regarding the positional accuracy of the toolset.
If there are any “Source of Truth” documents or technical blogs, we would greatly appreciate a pointer!
Thanks in advance,
Marco Radin
Hello @Marco_Radin,
I think we have some information around that floating around however it depends on the journey the data took from creation to tiling, and which tilers were used to convert it. For your built environment demos, what kind of data are you showcasing? Are you converting in Cesium ion with the Design Tiler.
Thank you,
Ben
Thanks for the reply, Ben.
To clarify our workflow: we aren’t using the tiling pipeline ourselves. Instead, we pull terrain data directly from Cesium Ion via the Google Maps Platform API.
In our scene, we overlay**Zmap data** (high-precision 3D models with centimeter-level accuracy), which serves as our ‘ground truth’ positioning framework. While we primarily use Cesium as a backdrop—and recognize that its resolution in the UK is roughly 2m—we are looking for a way to quantify the vertical and horizontal displacement between the Cesium terrain and reality.
Is there a specific metadata parameter, geometric error calculation, or known Root Mean Square Error we should be looking at to estimate this delta? We’d like to provide our users with a reliable ‘margin of error’ for measurements taken against the Cesium backdrop.
Hey @Marco_Radin ,
To clarify, are you accessing “Google Photorealistic 3D Tiles” through Cesium for Unreal (and thus cesium ion)? Or are you accessing directly via Google Maps Platform API?
Either way, this seems to be a question about the accuracy of Google Photorealistic 3D Tiles. From what I can find:
“Danbi: The Photorealistic 3D Tiles are not considered ‘survey-grade’ at this time. The original intent of the product is to power immersive visualizations, rather than support spatial analysis or model measurements. Also note that programmatically reading and recording measurements (heights, distances, elevations, etc.) from our 3D imagery is considered derivative and is prohibited. For details, in addition to our Terms of Service, please read our Policies page on “Pre-Fetching, Caching, or Storage of Content” where we explicitly state that any “image analysis, machine interpretation, object detection/identification, geodata extraction or resale, offline uses, including any of the above” based off of our imagery data is not allowed.”
Considering the sheer size of these datasets and being subject to a variety of survey methods and/or production methods, and that in the survey/geospatial area there are lots of different versions of “reality”, this estimate probably can’t be provided by us or google.
One could measure it against your definition of reality i.e. bring in context models that are accurate and georeferenced correctly for your use case and measure from there to the matching point on the google tiles, but I would note that according to the above excerpt from the Google Maps Platform FAQ: “programmatically reading and recording measurements (heights, distances, elevations, etc.) from our 3D imagery is considered derivative and is prohibited”. Whether this includes or excludes user made measurements or algorithmic based surveying… I can’t be sure.
I will dig around and see if the above info on accuracy has changed since that FAQ was made, or if I can find a quantified ± number, but this would also be affected by the LOD of the 3D tile loaded in that you take the measurement against, and come back here if I find anything.
Hi @Marco_Radin,
Unfortunately Google does not release the kind of information you’re looking for with Google Photorealistic 3D Tiles. It’s a bit of a black box when it comes to data accuracy and dates when it was captured.