Question on Multi-Sensor Calibration and Timestamps in RecFusion SDK


I am currently exploring the capabilities of the RecFusion SDK for a project that involves multi-sensor reconstruction. The user guide provided information on how to perform multi-sensor reconstruction using the RecFusion Pro software. However, I couldn’t find specific documentation examples related to achieving multi-sensor calibration using the RecFusion SDK.

Upon reviewing the SDK samples, specifically QtReconstruction and RealSenseReconstruction, I found that they come close to addressing this requirement. However, I would greatly appreciate confirmation from the community or the RecFusion development team regarding the implementation of multi-sensor calibration with the RecFusion SDK.

Additionally, I have a question regarding multi-view reconstruction. After capturing a sequence, I receive the message “[RGBDMultiSequence] Timestamps of at least one sequence are all zero. Frames will be matched by position instead of timestamps.” Even though I implemented threading to synchronize the sensors, the message persists. My question is whether this message warns about the impact on the quality of the resulting mesh, or if it can be safely ignored.

I would greatly appreciate any insights, guidance, or information you can offer on these matters. Thank you for your assistance!

The SDK sample MultiViewReconstruction contains sample code showing how to use the calibration from the SDK in the method MainWindow::performCalibration.

The reason you receive the message about the timestamps is that in the RecFusion SDK we do not store timestamps in the sequences, so we cannot use them for matching across different sequences. How do you actually load the sequences to get that message? Is that from the SDK or RecFusion Pro? If you already synchronize your cameras and you are sure that frame n in sequence i was taken at the same time as frame n in sequence j, then you can safely ignore the warning.