The method developed during this internship uses OpenVino, basic image processing methods, and some math, and gives us an enhanced image without the haze. The video below shows the sample case where this is implemented.
On the left, we can see an enhanced video feed. The right one shows the original image captured by an OAK-D Lite camera. The research spans from basic image processing techniques to advanced deep learning models using CNN and PyTorch.
Some more results are shown below where the UIEB Dataset was used for the images:
The outputs on the right seem to be decent enough to be utilized for further processing.