How to detect cars in road that never move in complete video?

Question

I used background subtraction to detect moving cars,and also to detect stopped cars on the road,but to accurately find non moving cars is tedious. I've tried cascade classifiers,but it gives too much false positives,and it would be helpful,if a can separate the regions as moving and non moving dynamically.

Answer 1

A cascade classifier works on an image, where as whether a car is moving or not comes from info across frames (time). Could you add details on how exactly you tried using the cascade? A cascade can possibly detect cars, but you would need another layer of intelligence/rules on top of it to infer whether it was moving or not.

Are you getting lot of falses while running the cascade on every frame? If so, some things you could try:

1. Modify cascade training params to make things tighter (more false rejection) at each stage
2. Add more layers - i.e. train longer. if the cascade training is running out negatives for higher layers, you may want to add a lot more images from where it can take negative samples.
3. Try correlating boxes detected across time as a way of reducing falses

After you have achieved a good enough detection rate with false positives low enough, you can focus on the logic needed to figure out whether a car is stopped or not, possibly based on tracking the detected car over time.