Mediaboss Marketing

There are a lot of options out there for scanning film these days, but theres just something about building a device yourself. This one from Benjamin Bezine does so using Lego and a Raspberry Pi. What makes this solution a little special, though, is that it uses machine learning and vision AI in order to determine the edges of frames, so you dont need to sit and operate it manually or worry about the number of turns changing as roll thickens and thins out or counting sprockets.

Its called RoboScan, and Benjamins been working on it for a while now (its his lockdown project) but hes not finished yet! Its an open-source project and hes been uploading the code to GitHub. Only 80 images were used to train the Raspberry Pi so that it knows what to look for, but it seems that its very effective with just that limited set.

RoboScan uses a backlight behind the film in order to provide even illumination across the frame, and when it detects that a complete frame is in the correct position, it tells the DSLR or mirrorless camera to which its attached to shoot a photo. The camera you choose to use should be compatible with libgphoto2, but that is a large list of cameras.

The first version was a proof of concept, Benjamin writes, but was far too sensitive, imprecise and had too many manual operations that could introduce user error. So now it uses machine learning to detect when a photo is correctly framed in front of the backlight using a Google Coral TPU. The whole thing is controlled by the web, and youre even able to add metadata.

Its a very cool project and a great use of Lego and AI. Fantastic if youve got a whole load of film you need to scan in and dont want to have to do it manually!

If you want to have a go at building your own or just find out more about RoboScan, you can find the full Lego model with a complete parts list over on Mecabricks and the source code can be downloaded from GitHub.

[via Raspberry Pi]

Read the rest here:
Automating scanning film with the help of Lego, a Raspberry Pi and a little machine learning - DIYphotography