The goal of pycambot is to replace a human camera operator with an autonomous pan-tilt-zoom camera. The application it was designed for is live streaming of a human speaker, with the speaker moving against a static background, and no other subjects in motion.
The bot uses OpenCV native face detection to identify the subject. It will revert to a safe PTZ position if it loses the subject's face for a few seconds.
The only camera tested with pycambot so far is the PTZOptics 20xUSB. The solution was tested on a UDOO x86 SOC board, an Intel all-in-one board with the processing power to capture HD video via a USB 3.0 interface and run it through OpenCV libraries at a reasonable frame rate.
If you are new to OpenCV, save yourself some time and use Docker images for kickstarting your environment. I recommend running Docker on Ubuntu and using one of these images. But first read the excellent tutorial from whose steps the Docker image was constructed.
- Python 2.7
- OpenCV 3.2
- HD PTZ camera supporting VISCA over serial and USB 3.0 HD video transfer