Focusing on a Pi, Kinect and DSLR for the Perfect Video
A DSLR camera for video
The Canon 5Dmk2 was a game changer for independent video makers. It delivered a stunning footage quality at a fraction of the cost of a professional camera. The shallow depth of field of a full frame DSLR camera is a strong asset, but keeping your subject in focus is a challenge. Even now, 8 years later my Canon 6D still cannot keep a subject in focus during movie recording. You need to focus manually, which is very very hard with a shallow depth of field.
Since I bought my Canon 5Dmk2 8 years ago, I did several attempt to design a focus follower that could keep the subject in focus. They were all based on ultrasound distance sensors, but these sensors were not sufficiently reliable.
With 3D sensing technology evolving in a fast pace (lidar becoming affordable due to the developments in autonomous vehicles, TOF (time of flight) sensors, Intel Realsense, ...) I felt it was time to give it a new try.
I had a first generation Kinect available which proofed to be an extraordinary piece of hardware in earlier project. It creates a very accurate depth image of the scene, with a depth up to 6 meters. Sufficient for my proof of concept.
Here is a video demonstrating the proof of concept. Notice the region of interest that is selected (on the metallic statue or on the wooden statue) and how the focus is kept accordingly.
Project Information URL: http://kinectfocuspuller.weebly.com/
Project [Make Your Own] URL: http://kinectfocuspuller.weebly.com/make-your-own.html