- RaspberyPi 4B
- Ubuntu 20.04 / ROS2 galactic
- StereoCamera : Arducam 4Channel Adapeter + two 8MP Rpicamera V2
- yaboom robot
I planned mobile robot that moves around my home. It will construct with yaboom robot H/W & stereo camera. yaboom robot will use for moving parts that controled by I2C and one Ultrasonic sensor. other original parts of yaboom will be remove. stereo camera will be made by arducam's 4 Channel multiplexer and two 8MP raspberry camera V2 and will make it produce a disprity Image based on AnyNet or RTSSNet CNN Network. after all things are built, I will try S-PTAM algorithms.
- psn : false.
ros2 run image_view disparity_image --ros-args -r /image:=disparity
apt -get install -y libopencv-dev \
libboost-all-dev \
pip3 install easydict, opencv-python
pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu
AnyNet is CNN network. and it's implementation is based on CUDA.
so, I modify some coded to run rasberrypi's single CPU and it added as submodule.
cpu version : AnyNet
Test on Colab AnyNet : Colab Test
GPU pretrained model dict's keys are have 'module.' prefix. so, need to remove it for load on single CPU.
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in checkpoint['state_dict'].items():
name = k.replace("module.", "") # remove `module.`
new_state_dict[name] = v
model.load_state_dict(new_state_dict, strict=False)
git clone https://github.com/WiringPi/WiringPi.git && cd WiringPi/ && \
./build && cd .. && rm -rf WiringPi/
apt-get install -y python3-rosdep \
ros-galactic-camera-calibration-parsers \
ros-galactic-image-common \
ros-galactic-image-pipeline \
ros-galactic-image-geometry
git clone https://github.com/Miragecore/cando.git
cd cando
git submodule update --init --recursive
colcon build
source /opt/ros/galactic/setup.sh
#move to cloned directory
. install/localsetup.sh
ros2 launch pipeline_launch pipeline_launch.launch.py