Overview / Usage

This project aims to solve the problem of lack of Visibility due to Smog and increasing Accidents. Inadequate visibility is an important factor that influences the risk of a road crash among all types of road user. Our solution is to build Intelligent Vision System that can detect obstacles in less visibility(i.e. Smog/Fog) using state of the art technologies like YOLO(You only look once), to detect the exact location of the object and thus mark the contours on the image along with the name of the object.
Using stereocameras can further help to calculate the distance of the obstacle.

Methodology / Approach

The initial approach which was to be used for solving this problem statement was to use a saliency map of the image and process it separately using YOLO, and combine the results obtained to remove the false positives from the image.
This was done by using two two dimensional array and plotting the confidence value of every pixel in the image obtained from both the processes and merging the results using Bayes theorem.
Another approach we are trying as of now is training a CNN using a foggy image dataset, and then merging those results with YOLO.