开源SLAM：复现DynaSLAM完整流程

开源SLAM——复现DynaSLAM

• 环境配置
• 配置ORB-SLAM2 配置Mask RCNN 配置pycocotools 配置DynaSLAM 文件检查修改 测试 问题

1. 论文
   
       Berta B , Facil J M , Javier C , et al. DynaSLAM: Tracking, Mapping and Inpainting in Dynamic Scenes[J]. IEEE Robotics & Automation Letters, 2018:1-1.
2. 源码
   https://github.com/BertaBescos/DynaSLAM

（搞了半天没跑通，待更…

环境配置

配置 ORB-SLAM2

C++11 or C++0x Compiler, Pangolin, OpenCV and Eigen3

配置Mask RCNN

sudo apt-get install libboost-all-dev.
sudo pip2 install keras==2.0.8
sudo pip2 install tensorflow==1.3.0
sudo pip2 install numpy==1.15.2

# get Mask_RCNN
git clone https://github.com/matterport/Mask_RCNN.git
cd Mask_RCNN

# delete keras & tensorflow in requirements.txt and avoid to install them again.
# while failed
sudo pip2 install -r requirements.txt	# failed

# replaced by below
sudo pip2 install scipy
sudo pip2 install Pillow
sudo pip2 install cython
sudo pip2 install matplotlib
sudo pip2 install scikit-image	# this step will reinstall Pillow & matplotlib
sudo pip2 install opencv-python # failed
sudo pip2 install h5py
sudo pip2 install imgaug		# failed because of opencv-python
sudo pip2 install IPython

# reinstall opencv-python
# commond "sudo pip2 install opencv-python" will install latest version,
# while latest opencv dont surport python2.7, so...
pip install opencv-python==4.2.0.32 

# reinstall imgaug
sudo pip2 install Shapely
sudo pip2 install imgaug

# install Mask_RCNN
sudo python2 setup.py install

配置 pycocotools

git clone https://github.com/waleedka/coco.git
cd coco/PythonAPI
make
sudo make install

配置DynaSLAM

git clone https://github.com/BertaBescos/DynaSLAM.git
cd DynaSLAM
	
# dowmload mask_rcnn_coco.h5 model: https://github.com/matterport/Mask_RCNN/releases.
# Place the model in the folder DynaSLAM/src/python/.
# copy /coco/PythonAPI/pycocotools to /Dynaslam/src/python  
# check python files & check cpp files & check cmakelists.txt 

chmod +x build.sh
./build.sh

文件检查修改

# check python files
# edit /Dynaslam/src/python/Check.py    
ROOT_DIR = "./" 
python Check.py # check
# then you can see "Mask R-CNN is correctly working"

# check cpp files 
# edit /Dynaslam/src/viewer.cc
# Comment "cv::imshow("DynaSLAM: Current Frame",im);"
# replaced by below
if(!im.empty())
{
   cv::imshow("DynaSLAM: Current Frame",im);
}

# check cmakelists.txt 
# Comment the code about mono_carla

测试

# Download a sequence from http://vision.in.tum.de/data/datasets/rgbd-dataset/download and uncompress it.
# place the sequence into /data and uncompress it. /date, which located in the same dir with /Examples, will be built by ourself.
# Download associate.py
python associate.py PATH_TO_SEQUENCE/rgb.txt PATH_TO_SEQUENCE/depth.txt > associations.txt
# associations.txt is same as below
./Examples/RGB-D/rgbd_tum Vocabulary/ORBvoc.txt ./Examples/RGB-D/TUM3.yaml ./data/rgbd_dataset_freiburg3_walking_xyz ./data/rgbd_dataset_freiburg3_walking_xyz/associations.txt ./data/mask ./data/output

问题

# This program requires version 3.13.0 of the Protocol Buffer runtime library, but the installed version is 2.6.1.
# ref: https://github.com/BertaBescos/DynaSLAM/issues/16#issuecomment-477580621
# commod: ./Examples/RGB-D/rgbd_tum Vocabulary/ORBvoc.txt ./Examples/RGB-D/TUM3.yaml ./data/rgbd_dataset_freiburg3_walking_xyz ./data/rgbd_dataset_freiburg3_walking_xyz/associations.txt



免责声明：本文系网络转载或改编，未找到原创作者，版权归原作者所有。如涉及版权，请联系删