Background

The purpose of this repo is 2-fold. On the one hand I am providing rough guidance for someone to make a stereo camera using ros2.

• I am not providing a fully detailed set of instructions because there are blogs which have dedicated pages to such instructions. On the other hand this repo represents my "scratching an itch" to empirically compare the effect of rotating cameras towards the focal point of interest. In this example I rotate the cameras inwards nby 0.6 degrees in order to have them focus at a theoretical center point which is 6-8 meters away. The cameras will have a base line distance of 82 mm.

In order to fully compare the effect of rotating the cameras I simultaneously set up another stereo camera with cameras that also have an 82mm baseline distance. These cameras are facing forward with no rotation.

The cameras were attained via amazon and can be found at :

• https://www.amazon.com/ELP-Megapixel-Distortion-Webcamera-Industrial/dp/B0BGH2LDMP

The housing for the cameras is 3d printed and made on fusion 360.

• Housing

NOTE :

• I've decided to use tmuxp in honor of NASA's LAMP stack being launched like that (at time of writing). If you don't have tmuxp then I suggest you get it so that you don't have to open a slew of various terminals. Eventually a launch file could be made in lieu of the tmuxp config file... but then that may require building a package simply to launch.
  • https://github.com/NeBula-Autonomy/LAMP/tree/6caea1b69f95b3288c6cdb2279cc96ac2f461bd3

Set Up

Get the repo

cd ~/workspace
git clone [email protected]:1hada/custom-depth-camera.git --recurse-submodules

Environment Configurations

cd ~/workspace/custom-depth-camera

mkdir -p ~/.tmuxp 
mkdir -p ~/.ros/camera_info
sudo mkdir -p /etc/udev/rules.d/ 

# I use a USB hub dedicated to these cameras and define these rules to make it as easy as pie to know exactly which camera corresponds to left/right and upper/lower
# NOTE : if the symlinks to the /dev/camera_* devices is not pointed toward the video devices then the vl42 package will not be able to run the camera. This can be resolvved by running `sudo udevadm control --reload-rules & sudo udevadm trigger `
sudo cp .udev/99-usb-camera.rules /etc/udev/rules.d/99-usb-camera.rules

cp .tmuxp/*  ~/.tmuxp/
cp .camera_info/*  ~/.ros/camera_info/

source /opt/ros/humble/setup.bash
cd ~/workspace/custom-depth-camera
colcon build

Get Camera Calibrations

In order to get the camera calibration parameters you'll need to print a calibration pattern. For my example I use a checkerboard pattern with the following specifications --size 12x8 --square 0.028

tmuxp load ~/.tmuxp/get-camera-config.yaml

If everything is connected correctly you will be able to press "calibrate" and then "save" the calibration file.

• Be sure to also rename the calibration file if you are like be and have mutlple stereo cameras. http://wiki.ros.org/camera_calibration/Tutorials/StereoCalibration

Run 2 Stereo Cameras

tmuxp load ~/.tmuxp/double-camera-to-depth-suite.yaml

Results

During this experiment I printed slots to hold 4 cameras.

• There were two "top" cameras used as a pair for the "top" stereoscopic camera.
  • These cameras were rotated towards eachother by 0.6 degrees (0.010472 radians)
• There were two "bottom" cameras used as a pair for the "bottom" stereoscopic camera.
  • These cameras are facing straight.

The results of this experiment are emperical since they are meant to determine further research into the practical effect of camera angles for stereo cameras which have a wide field of view. The results show that the any variance in the camera angle is negligible and may in fact result in a worse point cloud (due to angle accuracy as per any manufacturing tolerances) than the simpler set up of simply have the cameras face forward.

Here is a video showing the resulting point clouds. Yellow is the Lower stereo camera and White is the top stereo camera where the cameras are angled inwards.