DEEPFOG:AI

Inspiration

Improving autonomous cars reactions to difficult visual conditions during bad weather.

What it does

Using LIDAR, we classify the current visual conditions based on density of fog on the road.

How we built it

First we normalize the data by removing all unnecessary points from the dataset. We transform that 3d point cloud to a 2d density map. Then we classify the conditions based on the density map using a CNN.

Challenges I ran into

The dataset included large amounts of irrelevant data. We had to find an approach to extract the relevant information and downsample the extensive data set. Because the data was only stored on one local hard drive, we had to manually copy and process it locally on our notebooks. This lead to mayor delays while copying the data.

Accomplishments that we're proud of

By downsampling the data we managed to transform the complex problem into a smaller one, that can be solved at runtime event by on board hardware. In addition to that, we achieved high accuracies on our dataset.