This forum is disabled, please visit https://forum.opencv.org
 | 1 | initial version |
Since car windshields have quit a versatile shape, I would suggest to go for YOLO. My guess is that the cascade classifier is not versatile enough and works better for a more rigid object. If you want to stick with classical methods, and not deep learning, have a look at ICF/ACF/DeformableParts, but they have no integrated implementation in OpenCV.
Also consider the SSD detector, which is also a very efficient implementation. If you want to go a step further (and move outside OpenCV, then have a look at the Mask-RCNN detector en segmentor, which will give you the exact window area within your detection, and that in a single run through the architecture.
