We have currently a autonomous car in a university project based on the Nvidia Xavier. The camera preprocessing routine is written in Python with OpenCV.
In the Camera-Routine the fisheye lense of the camera is calibrated and a binary threshold is applied on the images of the camera for lanedetection.
We now wanted to use the Nvidia VPI with C++ to make the Preprocessing-Routine more efficient. Would a use of VPI with C++ be significantly faster than opencv with python?
One more problem is, that the Nvidia VPI doesn’t serve a threshold function. Would be a threshold implementation possible in a custom vpi function or isn’t that worth and we should use the OpenCV Threshold function for that?
Would it be efficient to to the calibration of the fisheye lens with nvidia vpi and then afterwards the threshold with opencv, or is it better to just stick to opencv with python and dont change anything at all?