<!DOCTYPE html>
<html>
<head>
GNSS Integration Email
</head>
<body style="font-family: Arial, Helvetica, sans-serif; color:#222; line-height:1.5;">
Hi Mathieu,
I’m working on an iPhone Pro application that captures the environment using the phone’s camera, LiDAR, IMU, and an external GNSS-RTK antenna.
The goal is to generate a geo-referenced 3D textured mesh and point cloud with high positional accuracy.
Because our geo-location requirements are strict, we use an external RTK antenna that theoretically provides centimeter-level accuracy.
I would like to integrate this GNSS-RTK data directly into RTAB-Map’s graph optimizer—not only to assist with loop closures, but also to
anchor the reconstructed map in an absolute geo-referenced frame and improve the overall optimization.
From what I understood, we add a new graph node with:
rtabmap::SensorData data(rgbImage, depthImage, id, timestamp);
rtabmap.process(data);
And then I attach the available IMU and GPS information:
data.setGPS(gps);
data.setIMU(imu);
Challenge: Sensor Rate Mismatch
Since each sensor has a different capture rate, the RGB, depth, IMU, and GNSS timestamps are not perfectly aligned.
If I simply attach the latest GNSS reading to each SensorData object, the GNSS pose may be off by several centimeters,
which would degrade the accuracy we expect from RTK.
Solution 1: Interpolation
One approach would be:
- Cache
SensorData objects until GNSS measurements before/after their timestamps are available.
- Interpolate the GNSS pose to match the exact timestamp of the RGB/depth frames.
- Optionally create separate nodes for RGB and depth frames, each with their own interpolated GNSS pose.
Would this be an appropriate strategy, or unnecessarily complex?
Solution 2: Adding GNSS Constraints Manually
Another option would be to add GNSS constraints directly into the graph:
Link link(fromId, toId, Link::kGlobalClosure, gpsPose, informationMatrix);
rtabmap.addLink(link);
My interpretation of the parameters:
- fromId: ID of the node for which the GPS measurement applies
- toId: Should this be
0 (world frame) or the same node ID?
- linkType:
Link::kGlobalClosure
- gpsPose: GNSS-derived pose
- informationMatrix: uncertainty of the GNSS measurement
This seems functionally similar to Solution 1, but allows more precise uncertainty modeling.
Is this the recommended method?
Graph Structure Question
From my understanding, each RTAB-Map node is tied to a SensorData object.
Is it possible to create a node containing only a pose and timestamp (without images)?
Or must every node be associated with an image/depth frame?
Main Question
Is this the correct way to integrate high-accuracy GNSS-RTK data into RTAB-Map’s optimization pipeline?
Thanks a lot for your help,
Julien Pennecot
</body>
</html>
Locked
