search

introlab / rtabmap_ros

RTAB-Map's ROS package.
http://wiki.ros.org/rtabmap_ros
BSD 3-Clause "New" or "Revised" License
934 stars 552 forks source link

Build with ORB SLAM2 #363

Open yubaoliu opened 4 years ago

yubaoliu commented 4 years ago

Hi,

I try to build with ORB SLAM2, I see this error:

[ 82%] Generating Javascript code from rtabmap_ros/SetLabel.srv [ 82%] Generating EusLisp code from rtabmap_ros/Point3f.msg /usr/bin/ld: cannot find -lpangolin collect2: error: ld returned 1 exit status rtabmap_ros/CMakeFiles/rtabmap_rgbd_sync.dir/build.make:1222: recipe for target '/home/yubao/data/catkin_ws/devel/lib/rtabmap_ros/rgbd_sync' failed make[2]: *** [/home/yubao/data/catkin_ws/devel/lib/rtabmap_ros/rgbd_sync] Error 1 CMakeFiles/Makefile2:16659: recipe for target 'rtabmap_ros/CMakeFiles/rtabmap_rgbd_sync.dir/all' failed make[1]: *** [rtabmap_ros/CMakeFiles/rtabmap_rgbd_sync.dir/all] Error 2 make[1]: *** Waiting for unfinished jobs....

Here is my flow:

  1. Build Pangolin standalone and add to library link path export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/data/software/Pangolin/lib

export Pangolin_DIR="$HOME/data/software/Pangolin/lib/cmake/Pangolin"

  1. Build ORB SLAM2 succeed
  2. Set ORB ROOT PATH export ORB_SLAM2_ROOT_DIR=/home/yubao/data/catkin_ws/src/ORB_SLAM2

image

image

  1. Build Rtabmap sucessfully

  2. Build rtabmap_ros under my ROS workspace use catkin_make, then I got that error. (Note: it succeeds without build ORB SLAM2 option)

Thanks

matlabbe commented 4 years ago

On my system with ORB_SLAM2, I make install Pangolin, so I don't need to modify LD_LIBRARY_PATH for it. You may look at this docker file which shows how to install ORB_SLAM2: https://github.com/introlab/rtabmap/blob/d4cb623d9ca522229bfeff4766bf8e22f9a9c17d/docker/jfr2018/Dockerfile#L162-L170

yubaoliu commented 4 years ago

Thanks, But I don't want to install Pangolin to my system path. Actually I can build ORB_SLAM2 standalone successfully. The problem happens when I build ORB_SLAM2 in Rtabmap/Rtabmap_ros. I set the Cmake DIR path and set the library link path LD_LIBRARY_PATH. However, Rtabmap_ros still cannot find it when compiling using catkin_make.

matlabbe commented 4 years ago

The -lpangolin flag would be coming from imported dependencies of rtabmap cmake config file. You can try to remove it from the installed RTABMapConfig.cmake, then build rtabmap_ros again. If it is building, we may look to remove this dependency flag from here.

zhuoyan28 commented 4 months ago

Can someone tell 1, RTAB-MAP commands to run KITTI image dataset alone, such as binocular images? Where will the output be stored in the ros environment?

  1. ORBSLAM+RTAB-MAP running KITTI image dataset, such as binocular image command? Where will the output be stored in ros environment?
matlabbe commented 4 months ago

  1. For the kitti image dataset, you could process it with the command line tool rtabmap-kitti_dataset (type in terminal alone and you will see usage and an example at the end). It is independent of ros. The database could be use in ros afterwards if needed, though I don't think there is a reason to do so with kitti dataset (unless it was your own dataset where a robot would navigate in afterwards).

  2. The command would look like:

    
rtabmap-kitti_dataset \
   --Rtabmap/PublishRAMUsage true\
   --Rtabmap/DetectionRate 2\
   --Rtabmap/CreateIntermediateNodes true\
   --RGBD/LinearUpdate 0\
   --GFTT/QualityLevel 0.01\
   --GFTT/MinDistance 7\
   --Odom/Strategy 5\
   --OdomORBSLAM2/VocPath ~/ORB_SLAM2/vocabulary/ORBvoc.txt\
   --OdomORBSLAM2/ThDepth 40\
   --OdomORBSLAM2/Fps 10\
   --OdomORBSLAM2/MaxFeatures 2000
   --Mem/STMSize 30\
   --Kp/MaxFeatures 750\
   --Vis/MaxFeatures 1500\
   --gt "~/KITTI/devkit/cpp/data/odometry/poses/07.txt"\
   ~/KITTI/dataset/sequences/07


Note that if you want to reproduce the results of the [paper](https://introlab.3it.usherbrooke.ca/mediawiki-introlab/images/7/7a/Labbe18JFR_preprint.pdf) (with same parameters), see https://github.com/introlab/rtabmap/tree/master/docker/jfr2018
zhuoyan28 commented 4 months ago

really appreciate your patience! Thanks.

---- Replied Message ---- | From | @.> | | Date | 03/25/2024 09:05 | | To | introlab/rtabmap_ros @.> | | Cc | Dan @.>, Comment @.> | | Subject | Re: [introlab/rtabmap_ros] Build with ORB SLAM2 (#363) |

For the kitti image dataset, you could process it with the command line tool rtabmap-kitti_dataset (type in terminal alone and you will see usage and an example at the end). It is independent of ros. The database could be use in ros afterwards if needed, though I don't think there is a reason to do so with kitti dataset (unless it was your own dataset where a robot would navigate in afterwards).

The command would look like:

rtabmap-kitti_dataset \ --Rtabmap/PublishRAMUsage true\ --Rtabmap/DetectionRate 2\ --Rtabmap/CreateIntermediateNodes true\ --RGBD/LinearUpdate 0\ --GFTT/QualityLevel 0.01\ --GFTT/MinDistance 7\ --Odom/Strategy 5\ --OdomORBSLAM2/VocPath ~/ORB_SLAM2/vocabulary/ORBvoc.txt\ --OdomORBSLAM2/ThDepth 40\ --OdomORBSLAM2/Fps 10\ --OdomORBSLAM2/MaxFeatures 2000 --Mem/STMSize 30\ --Kp/MaxFeatures 750\ --Vis/MaxFeatures 1500\ --gt "~/KITTI/devkit/cpp/data/odometry/poses/07.txt"\ ~/KITTI/dataset/sequences/07

Note that if you want to reproduce the results of the paper (with same parameters), see https://github.com/introlab/rtabmap/tree/master/docker/jfr2018

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

zhuoyan28 commented 3 months ago

install dockerfile cmd: $ cd rtabmap $ docker build -t introlab3it/rtabmap:jfr2018 -f docker/jfr2018/latest/Dockerfile . Dear author, three questions please 1、According to the dockerfile of jfr2018(not ros version), the file run_all.sh only keeps rtab-map(f2m) with orb2-rtab(os2) and the results can be reproduced.

RTAB-Map version: 0.20.23 ----I installed， but the RTAB-Map version in docker is 0.16.3 ~/study_learning/rtabmap/docker/jfr2018$ ./run_all.sh (only 04 file) rtabmap-kitti_dataset [ WARN] (2024-04-09 20:42:36.688) Parameters.cpp:1135::parseArguments() Parameter migration from "OdomORBSLAM2/VocPath" to "OdomORBSLAM/VocPath" (value=/root/ORBvoc.txt). [ WARN] (2024-04-09 20:42:36.688) Parameters.cpp:1135::parseArguments() Parameter migration from "OdomORBSLAM2/ThDepth" to "OdomORBSLAM/ThDepth" (value=40). [ WARN] (2024-04-09 20:42:36.688) Parameters.cpp:1135::parseArguments() Parameter migration from "OdomORBSLAM2/Fps" to "OdomORBSLAM/Fps" (value=10). [ WARN] (2024-04-09 20:42:36.688) Parameters.cpp:1135::parseArguments() Parameter migration from "OdomORBSLAM2/MaxFeatures" to "OdomORBSLAM/MaxFeatures" (value=2000). Paths: Sequence number: 04 Sequence path: ./datasets/kitti/dataset/sequences/04 Output: ./results/kitti/04 Output name: orbslam2 left images: ./datasets/kitti/dataset/sequences/04/image_0 right images: ./datasets/kitti/dataset/sequences/04/image_1 calib.txt: ./datasets/kitti/dataset/sequences/04/calib.txt times.txt: ./datasets/kitti/dataset/sequences/04/times.txt Ground Truth: ./datasets/kitti/devkit/cpp/data/odometry/poses/04.txt Exposure Compensation: false Disparity: false Saved calibration "orbslam2_calib" to "./results/kitti/04" Parameters: FAST/Threshold=12 GFTT/MinDistance=7 GFTT/QualityLevel=0.01 Kp/FlannRebalancingFactor=1.0 Kp/MaxFeatures=750 Mem/BinDataKept=false Mem/STMSize=30 Mem/UseOdomFeatures=false Odom/Strategy=5 OdomF2M/MaxSize=3000 OdomORBSLAM/Fps=10 OdomORBSLAM/MaxFeatures=2000 OdomORBSLAM/ThDepth=40 OdomORBSLAM/VocPath=/root/ORBvoc.txt OdomViso2/MatchNmsN=7 OdomViso2/MatchRadius=300 RGBD/AngularUpdate=0 RGBD/LinearUpdate=0 RGBD/ProximityBySpace=false Rtabmap/CreateIntermediateNodes=true Rtabmap/DetectionRate=2 Rtabmap/PublishRAMUsage=true Rtabmap/WorkingDirectory=./results/kitti/04 Vis/CorGuessWinSize=20 Vis/CorNNDR=0.6 Vis/MaxFeatures=1500 RTAB-Map version: 0.20.23 Processing 271 images... [ERROR] (2024-04-09 20:42:36.780) OdometryORBSLAM.cpp:1164::computeTransform() RTAB-Map is not built with ORB_SLAM support! Select another visual odometry approach. [ERROR] (2024-04-09 20:42:36.780) Rtabmap.cpp:1337::process() RGB-D SLAM mode is enabled, memory is incremental but no odometry is provided. Image 1 is ignored!

2, Then. I want to reproduce the result of stereo of ORBSLAM2 standalone , how can I do it? 3, I want to modify the code of rtab-map, how to run the modified code?

matlabbe commented 3 months ago

The jfr2018 docker is a snapshot of the code used for the paper. To use latest rtabmap, use docker under latest folder: https://github.com/introlab/rtabmap/tree/master/docker/jfr2018/latest

Normally, run_kitti_dataset.sh should use the right rtabmap built with orb-slam when using docker approach. Make sure to change the image name in case you rebuilt your own docker image: https://github.com/introlab/rtabmap/blob/4e7110930c594187104c52083f2cf469157276c1/docker/jfr2018/run_kitti_datasets.sh#L96

For "ORBSLAM2 standalone", do you mean using only ORBSLAM2 evaluation tools? I think they have one for kitti: https://github.com/raulmur/ORB_SLAM2/tree/master/Examples/Stereo

To do custom tests, either you change the code and rebuild the container, or build/install rtabmap on host machine with dependencies you need. You can use run_kitti_dataset.sh locally without docker container if you set DOCKER variable to 0.

zhuoyan28 commented 3 months ago

The algorithm system was installed by myself, not the docker file the command is : rtabmap-kitti_dataset \ --Rtabmap/PublishRAMUsage false\ --Rtabmap/DetectionRate 2\ --Rtabmap/TimeThr 0 \ --Rtabmap/MemoryThr 0\ --Rtabmap/LoopThr 0.11\ --GFTT/QualityLevel 0.01\ --GFTT/MinDistance 7\ --Odom/Strategy 0\ --OdomORBSLAM2/VocPath /home/shen/rtab_ws/src/ORB_SLAM2/Vocabulary/ORBvoc.txt\ --OdomORBSLAM2/ThDepth 40\ --OdomORBSLAM2/Fps 10\ --OdomORBSLAM2/MaxFeatures 3000\ --OdomF2M/MaxSize 3000 \ --Odom/KeyFrameThr 0.3\ --Mem/STMSize 30\ --Mem/RehearsalSimilarity 0.2\ --Kp/MaxFeatures 750\ --Vis/MaxFeatures 1500\ --Vis/CorNNDR 0.6\ --Vis/CorGuessWinSize 20\ --Vis/MinInliers 20\ --gt "/home/shen/study_learning/datasets/kitti/devkit/cpp/data/odometry/poses/04.txt"\ /home/shen/study_learning/datasets/kitti/dataset/sequences/04

but the result for kitti-04 is image

image How to reproduce your results？

matlabbe commented 2 months ago

To recover results from the paper, you should use rtabmap-report --scale:

rtabmap-report --scale --kitti results/kitti/04
Directory: results/kitti/04
   f2m.db (271, s=0.987):   error lin=0.214m (max=1.514m, odom=0.214m) ang=0.2deg, KITTI: t_err=0.45% r_err=0.13 deg/100m, slam: avg=40ms (max=46ms) loops=0, odom: avg=43ms (max=63ms), camera: avg=5ms, map=172MB
   os2.db (271, s=1.001):   error lin=0.158m (max=0.229m, odom=0.158m) ang=0.6deg, KITTI: t_err=0.41% r_err=0.16 deg/100m, slam: avg=41ms (max=49ms) loops=0, odom: avg=39ms (max=48ms), camera: avg=6ms, map=709MB

From the paper (section 4):

With the offline datasets, the different stereo odometry approaches tested are using their own stereo corre- spondence approaches, which can generate slightly different disparity values. We also observed that camera calibration are not always accurate: rectified images still contain some visible distortions. These problems can influence the scale of the created map, thus bias trajectory accuracy when comparing with the ground truth (for which we assume there is no scale error). As observed in [Mur-Artal and Tard´os, 2017], depth image values from some TUM RGB-D sequences can be slightly off, causing also a scale problem. For a fair comparison between all visual odometry approaches, as we cannot know if the scale problem is caused by the camera calibration and/or the disparity computation approach, results are presented with the map scaled to minimize the error against the ground truth.