I want to write a python script to highlight cars in a picture (with a reactangle (OPENCV)). I'm going to make a HAAR Cascade, but I'm a bit stuck. I read a lot about it, and I have some concerns.
I'm planning on using pictures of cars without a background (no road or anything like that) as my positive images. As negative images I would use empty roads.
Also I'd like to know how many pictures should I get? On forums I've read that I would need 1000 positive and 10 000 negative pictures.
Is my thought process correct?
HAAR cascade is old technology (before convolutional neural networks), nowadays there are more advanced techniques to solve car detection problems. The way you are approaching the problem is correct, the problem is data quality can greatly influence results. I did a demo using a more advanced algorithm if you want to check (it comes with pre-trained models).