Since the raspberry Pi 5 has two camera ports, and Picam modules are inexpensive, I am wondering if any one has successfully set up a stereo vision node using camera_ros or something similar?
To use two independent camera’s for stereo vision you would need:
- Synchronize capture start of the two sensors, which is typically done with hardware. The start capture pulse or command would need to be issued to the cameras with <50us jitter between each camera
- Software to keep the exposure, and white balance of each camera the same, programmed during the vertical blank when the sensor is not capturing data.
- A rigid assembly with mechanical tolerance for vibration, shock, and thermal changes, for extrinsic calibration between the camera’s that remains valid over some time interval.
- Calibration software for extrinisics between the two camera’s.
- Intrinsic + extrinsic calibration per camera.
- Sufficient compute or hardware acceleration for a stereo matching algorithm or AI perception function.
The effort involved may quickly exceed the cost savings of the cheap camera.
Thanks. It would seem that a better use of two cameras is to separate tasks,
perhaps one camera used for fiducial markers, and another used for Yolo object recognition.