MartyG-RealSense
commented
3 years ago Hi @rahulsharma11 Does the stability of your IMU results after launch improve if you add the commands below to the end of your roslaunch instruction:
global_time_enabled:=false unite_imu_method:=linear_interpolation
Some RealSense ROS users have found that they have better results if they use unite_imu_method:=copy instead of linear_interpolation.
Intel's D435i SLAM guide - which also uses roslaunch realsense2_camera opensource_tracking.launch - offers the following advice about losing track during movement:
The built-in IMU can only keep track for a very short time. Moving or turning too quickly will break the sequence of successful point cloud matches and will result in the system losing track. It could happen that the system will recover immediately if stopped moving but if not, the longer the time passed since the break, the farther away it will drift from the correct position.
The odds for recovery get very slim, very quickly. The parameters set in the launch file are most likely not ideal but this is a good starting point for calibrating.
The D435i SLAM guide can be found here:
https://github.com/IntelRealSense/realsense-ros/wiki/SLAM-with-D435i
A RealSense ROS user expanded upon this guide to add information about setting up localization with RTABMAP to aid recovery when tracking drifts.
https://shinkansan.github.io/2019-UGRP-DPoom/SLAM.html
In regard to 'point to point' distance measurement with a point cloud, the link below provides guidance about using RViz to measure between two selected points:
https://github.com/IntelRealSense/realsense-ros/issues/553
Similar point to point selection and measurement is also available in the 3D point cloud mode of the RealSenSe Viewer tool.
Regarding use of IMU in SLAM: as you pointed out, IMU-less cameras such as D435 can be used for SLAM in programs such as ORB-SLAM2. SLAM that makes use of IMU data can be referred to as Visual-Inertial Odemetry (VIO) and can utilize the pose (position and rotation angle) of the camera. There are methods that can be used to get the camera angle without an IMU though, such as a plane-fit algorithm (which is built into the SDK's Depth Quality Tool).
rahulsharma11
Required Info -
Camera Model | D435i Firmware Version | 05.12.12.100 Operating System & Version | Linux 20.04 Kernel Version (Linux Only) | 5.8.9-050809-generic Platform | laptop SDK Version | 2.43.0 Language | C Segment | others
Hi, I am using roslaunch command - "roslaunch realsense2_camera opensource_tracking.launch initial_reset:=true is_dense:=true".
I have 4 quick querries.
1) While running the app, rviz shows decent results. But after few seconds like 10-15, it freezes and terminal output is- [ INFO] [1617772858.545138390]: Odom: quality=520, std dev=0.007428m|0.038960rad, update time=0.534487s [ WARN] [1617772859.394979272]: Failed to meet update rate! Took 0.010023749000000000201 [ WARN] [1617772859.396603784]: Failed to meet update rate! Took 0.0084350780000000003672 [ WARN] [1617772860.715949553]: Failed to meet update rate! Took 0.0066743310000000003607 [ WARN] [1617772865.583072671]: Still waiting for data on topic /imu/data_raw...
Then it runs normally. This process happens continously. I checked with lower fps also. Any steps to resolve this?
2) When i rotate the camera to get the 3d pointcloud, sometime it hangs and i can see IMU drifting starts which sometimes get resolved automatically but sometime not. output terminal shows- [ WARN] (2021-04-07 10:59:16.934) OdometryF2M.cpp:557::computeTransform() Registration failed: "Not enough inliers 0/20 (matches=34) between -1 and 564" [ INFO] [1617773356.935316361]: Odom: quality=0, std dev=0.000000m|0.000000rad, update time=0.089860s [ WARN] (2021-04-07 10:59:17.058) OdometryF2M.cpp:557::computeTransform() Registration failed: "Not enough inliers 0/20 (matches=24) between -1 and 565" [ INFO] [1617773357.059841865]: Odom: quality=0, std dev=0.000000m|0.000000rad, update time=0.090116s [ WARN] (2021-04-07 10:59:17.186) OdometryF2M.cpp:557::computeTransform() Registration failed: "Not enough inliers 0/20 (matches=28) between -1 and 566" [ INFO] [1617773357.187354359]: Odom: quality=0, std dev=0.000000m|0.000000rad, update time=0.091475s [ERROR] (2021-04-07 10:59:17.190) Rtabmap.cpp:1147::process() RGB-D SLAM mode is enabled, memory is incremental but no odometry is provided. Image 970 is ignored!
Does that mean i need to rotate the camera very slowly? and if drift does not autocorrected , then i need to restart it everytime?
3) Is there any way i can get point to point distance in pointcloud? like i want to confirm that wall in front is at 2meter should be exactly there in pointcloud? i saw "show_center_depth.py", but i need to see in my desired point to point distance in pointcloud.
4) General querry is why IMU is required to get the camera orientation? like in ORB_SLAM, stereo pair is sufficient to get the camera orientation.
Thanks.