3-D Reconstruction of Clouds From Multiple Images

Background

The ability to monitor the clouds layer that covers the Earth has a critical effect on weather forecasting.

Today, huge resources are needed in order to produce an accurate weather forecast. Various systems are used for this purpose e.g. satellites, radars systems, weather balloons etc.

Together with the advantages of these systems, they also have disadvantages such as, high price, inability to see the clouds shape from Earth.

Project Goals

The project goal is to develop and implement an algorithm, which reconstructs the 3-D cloud layer from multiple images by using a cheap and available optical means, such as, regular cameras, Smartphone cameras etc.

Our solution

First of all, to simulate the cloud layer and the captured image we use a Radiative Transfer solver (called SHDOM). With this tool we produce a grid of one hundred cloud fisheye images, which simulate images captured from cameras installed on the ground. Then we set a target point at which we are going to focus on, after that we direct all the cameras to this point, as our goal is to get multiple images of cloud from several diverse angels. After this stage, we have to make a transformation from fisheye images to rectangular images by wrapping around the target point region:

Note that:

1. The dimensional window images are constant for all images.

2. Matching pixels by making a coordinate transformation.

Once we have synthesized the perspective images as shown below, we can use silhouette algorithm which convert the colorful images to binary ones i.e. white and black, last step is essential to allow using the ‘dinosaur’ algorithm, which reconstruct 3-D object from multiple silhouettes.

Initial state, all cameras are looking upward. Next step all cameras are looking at the same object, set as a target point.

Showing the transformation from fish eye image to rectangular one, by wrapping the region marked in red.

Silhouette algorithm converts colorful images to binary ones, which all clouds being white while sky converts to black. It is essential step to get use of ‘dinosaur’ algorithm, since it’s only defined on binary images.

Results

Conclusions

A new 3-D reconstruction cloud layer approach was presented. Its main advantage is its low cost, availability and ease of use.

Acknowledgements

First and foremost, we would like to send big thank to our supervisor of this project, Mr. Amit Aides for the valuable guidance and advices. His willingness to motivate us contributed tremendously to our project. We also would like to thank Johanan Erez and the VISL staff for providing us with a good environment and facilities to complete this project. Also we appreciate the support of the Minerva Ollendorff Center.