Vision module is a separate process that will provide point cloud, stereo correspondence, disparity map, etc. to the controller module.
The communication will be via network sockets for easier synchronization. Initially I've considered shared memory (higher performance), but this is much more difficult to manage robustly.
Incorporated libv4l2cam as robo::Camera, which is now decoupled from OpenCV.
Currently using two Logitech C920 HD Pro cameras on a powered USB 2.0 hub (Manhattan 7 port.)
Original libv4l2cam code is based on Giacomo Spigler and George Jordanov.
- OpenCV
- Implement the following settings in robo::Camera to avoid different focus values in two cameras:
#
# disable auto-focus
#
v4l2-ctl -d 0 -c focus_auto=0
v4l2-ctl -d 1 -c focus_auto=0
#
# actual focus value to be determined empirically?
#
v4l2-ctl -d 0 -c focus_absolute=0
v4l2-ctl -d 1 -c focus_absolute=0
-
Implement camera calibration support
-
Implement stereo processing
-
Connect to controller module and wait for commands.
-
Determine good resolution / frame rate settings for 2 x USB cameras (and a wifi USB eth card) to avoid saturating the USB bus.