Scanning the sky
Scanning the sky always turned out to be the most complicated thing to do. Just to mention a couple details, it needs to be done quick, in high resolution, has to have true dynamic range, correct colors and needs to be fully automated from shooting to processing to make it sustainable when capturing a big variety of skies.
Luckily technology evolved a lot since the first HDRI shooting rig we developed. In the past we have experimented with different solutions and setups.
There are a lot of different ways how you can approach the same problem. If you decide that instead of rotating one camera into different positions you would like to shoot with a multi-camera system, there are many options how you can calculate and come to a conclusion of how many cameras you need and what type of lenses to use.
In order to improve the speed of the captures a lot, in our case we decided to go with a 3 camera setup, where we need to use fish-eye lenses to cover a sphere and capture the sky.
With a setup like this we had to tackle another issue, which is the ND filters. They are mandatory to use as otherwise the Sun cannot be captured correctly, it would stay overexposed and it would just appear super bright. But with fish-eye lenses and their curved front element it was impossible to use high quality ND filters in the past. But fortunately as technology evolved now there is a solution for this problem as well.
In the past most high-end cameras used to be DSLR’s. Lenses for DSLR’s are calculated to have a relatively high distance from the camera sensor as the mirror needs it’s place between them. With recent development in the digital processing in cameras, Mirrorless System Cameras are now so good and high end that we can use them for shooting HDRI’s.
This let’s us use a combination of the two systems to get some extra benefits. By using special adapters for DSLR type lenses we can actually utilize the space where in previous cameras the mirror used to sit and install our ND filters in this position.
When building a complex system there are a lot of factors you need to pay attention to and keep in mind. The same thing goes to the different camera brands. You want to control everything via a USB connection and things need to work fast. We looked into a lot of variations, tested them, but for different reasons many of them were not suitable for the tasks we wanted to perform. Based on our research in the end we decided to go with a Canon system.
Besides the ability to properly control the cameras remotely also Canon is the only brand that offers a high resolution DSLM (EOS R5, 45MPX) and a zoomable high quality fish eye lens combination.
The zoom feature is important with the lens as this way we can shoot in a really specific focal length that produces the least amount of black space outside the image. This comes in handy to clearly define the overlap between each angle and to increase the final resolution of the capture.
Also with the variable focal length we have the flexibility to use the same lenses but with a different setting, change and capture either a 360 degree, full sphere HDRI or go only for the sky-hemisphere and shoot that in an even higher resolution.
With all this in mind at this point we had to find the optimal orientations for the 3 cameras:
Planning is key. Simulating what happens and how the image coverage will turn out was very important. In the image above two different options are shown. One for maximum coverage of a full sphere environment (right) and one for maximum resolution (left) while leaving out some parts below the cameras.
On the image below you can see the planning of what the different cameras should capture in a certain setup and how the stitching process should look like. The projections of every camera can be rotated and adjusted in an optimal way while changing the focal length to find the perfect coverage.
After purchasing, designing and 3D printing all the different elements for the rig and doing our first test shoots they quickly showed that the planning was worth it. The 3 camera rig head was fine from the very first test.