| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
Hello, Does hector_slam create only 2D maps or can it be used to created 3D models as well? My data comes from a Velodyne LiDAR and I need to create a 3D model of the object that is being scanned. I am looking for the right tool to use and was not sure if hector_slam could create 3D models. If it can't and you can recommend something else, please do so :) Thank you.
hector_slam performs localization only in 2D, but if you combine that information with IMU data (for roll and pitch) and additional data about the z position (from prior knowledge, odometry or other sensors), it can be used for 3D mapping. This approach is used for building a 3D octomap-based map via mapping with known poses using a RGB-D sensor with our autonomous USAR robots as visible in this video.
Depending on your accuracy requirements, sensor motion and scale of environment other solutions might be worth a look. See for instance this recent video made using a Velodyne HDL-64e. ethzasl_icp_mapping also uses libpointmatcher and you could try that.
Thank you Stefan. It is not immediately clear from the video the quality of the 3D model, the resolution, etc. but I will look into libpointmatcher and ethzasl_icp_mapping. I am specifically interested in mounting VLP-16 on a drone and scanning a small building with a couple cm or better resolution.
Hi, I'm doing something similar as you are up to. I first tried using lpm/ethzasl_icp_mapping but was disappointed by the high percentage of points that you have to filter out to get it running in realtime. I found configuring it quite hard. Please let me know if you're better at tweaking params :)
Also, I implemented a point to plane algo that doesnt rely on kd trees for surface normal approximation and src/tar point association. It seems to be _a_lot_ faster (>20iter, 100%pts) than lpm and doesnt deteriorate with map size . I'll try to write up the approach and release code end of next month
Our team is also working on 3D SLAM by modifying some existing 2D SLAM algorithm such as hector_slam, gmapping and MRPT icp-SLAM. We are not sure whether we can successfully modify hector_slam because it seems hard to add IMU data to it, neither is gmapping. Hope to see any of your progress :)
We've mainly focused on building a 2D SLAM system recently. However one of our teammates planned to convert 2D point cloud to 3D using data obtained by IMU. He will have this done in maybe one month. I will inform you as soon as we have a progress. And we are also curious about your idea :)
i was successful in generating a 3D map of the indoor environments (somewhat rough). I have used hector_localization and hector_mapping for localizing my robot in a 3D environment and octomapps for generating the maps. hector_localization is provided with imu messages and poseupdate topic from the hector_mapping node. Thanks to Stefan for providing the file which is linked in this answer
Can you please email me at ateator@jhfletcher.com ? I am interested in a setup much like yours and would like to see your cose / ask you questions.
I've generated fairly accurate 3D maps of indoor environments using hector_mapping, hector_imu_attitude_to_tf, and octomap_server using a laser scanner, ultrasonic range finder for altitude, and an IMU for orientation.
@Icehawk101, how do you use a range finder for measuring altitude? And yes, it would be great to see the maps you were able to generate.
I have the ulrasonic sensor facing down to determine the range between the prototype and the ground, then I built a node that takes the pose estimate from hector_mapping, adds the height as a z value for the pose, then publishes a transform from base_footprint->base_stabilized
Asked: 2016-02-16 13:22:12 -0500
Seen: 3,856 times
Last updated: Jun 28 '16
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
