This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA4VM: ROI based perspective transform in LDC

Part Number: TDA4VM


Hi, 

I am looking at possibilities of doing ROI based Perspective transform inside LDC. I have a set of 4 point(rectangle of unequal length eg: like a photo hanging on a wall with picture taken at 40degree from the wall) and i want to map this 4 points to a rectangle of desired size. I am able to create the transform matrix needed to change the perspective using Matlab. But i am doubtful about selecting these 4 point and mapping to a rectangular output. 
I would like to the possibilities around this. 

Regards,
Dinesh

  • In general, I would not recommend using the integer transform matrix of LDC for arbitrary perspective transform, because it is an integer implementation with many limitations.
    It is fine to use only for small perspective changes, e.g., changing angle of view by a few degrees.

  • So, the recommendation would be to go with GPU for such intense transformation? Is that correct.

  • Yes, GPU is better for dynamic use cases.

    LDC is better for relatively static use cases, i.e., when you are able to prepare the mesh LUTs offline and optimize LDC parameters in tuning tool.
    It is also OK for adding small run-time adjustment with the integer perspective transform matrix, i.e., run-time adjustment for camera calibration.

    If you have to re-generate mesh LDC LUTs at run-time frequently, you may not have enough time to do it or you may not be able to optimize LDC parameters for good performance or avoiding running out of LDC internal memory.