Odometry: prepareRGBFrameBase asserts !depth.empty()

[JoeHowse]

System Information

OpenCV version: 5.x, commit d499581

Detailed description

prepareRGBFrameBase calls prepareScaleDepth, which asserts !depth.empty(). This assertion appears to be spurious because prepareRGBFrameBase is associated with the case OdometryType::RGB. OdometryType::RGB (unlike OdometryType::DEPTH and OdometryType::RGB_DEPTH) should not require a depth frame.

Steps to reproduce

The bug can be reproduced with Python code similar to the following, for any BGR images:

od = cv2.Odometry(cv2.ODOMETRY_TYPE_RGB)
odFrame1 = cv2.OdometryFrame(image=bgrFrame1)
odFrame2 = cv2.OdometryFrame(image=bgrFrame2)
od.prepareFrames(odFrame1, odFrame2)

This throws odometry_functions.cpp:49: error: (-215:Assertion failed) !depth.empty().

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)

[asmorkalov]

@savuor please take a look.

[savuor]

@JoeHowse Thanks a lot again!

The bug is not in the code but in the documentation.

All Odometry algorithms require depth data for calculation. Depth data for both frames is required to establish pixel-to-pixel correspondences at each iteration.
The difference is in the objective function they minimize:

• OdometryType::RGB minimizes the difference in image pixels
• OdometryType::DEPTH minimizes point-to-plane error (see KinectFusion paper)
• OdometryType::RGB_DEPTH combines both

We need to fix the documentation: now it says that RGB requires image data only.