Skip to content

Latest commit

 

History

History
94 lines (67 loc) · 3.83 KB

visual_frontend.md

File metadata and controls

94 lines (67 loc) · 3.83 KB
Raw

Run the visual front-end

Remember to source the workspace first

source ~/svo_ws/devel/setup.bash

Example bags and launch files

Download the test bag file from here. Launch SVO node:

roslaunch svo_ros run_from_bag.launch cam_name:=svo_test_pinhole

This will also start RVIZ. Then play the bag file:

rosbag play airground_rig_s3_2013-03-18_21-38-48.bag

Now you should be able to observe the camera motion and sparse map in RVIZ.

You can also download another bag recorded with a fisheye camera. Then you need to change the following line in run_from_bag.launch:

<rosparam file="$(find svo_ros)/param/pinhole.yaml" />

to

<rosparam file="$(find svo_ros)/param/fisheye.yaml" />

And launch SVO via:

roslaunch svo_ros run_from_bag.launch cam_name:=bluefox_25000826_fisheye

Customize launch files

We provide several example launch files under svo_ros/launch, such as:

  • run_from_bag.launch: run SVO on an topic that publishes images
  • live_nodelet.launch: run svo as nodelet. These files can not be launched directly, since they depend on some private packages. But they can be used as an example for using svo with other nodes.

In the launch file, you have to specify the following for svo node/nodelet, as in run_from_bag.launch:

    <!-- Camera topic to subscribe to -->
    <param name="cam0_topic" value="camera/image_raw" type="str" />
    
    <!-- Camera calibration file -->
    <param name="calib_file" value="$(arg calib_file)" />
    
    <!--Parameters-->
    <rosparam file="$(find svo_ros)/param/pinhole.yaml" />

  Note that SVO also supports stereo cameras.

Parameter files

We provide two example parameter files under svo_ros/param:

  • pinhole.yaml: for relatively small field of view cameras
  • fisheye.yaml: for cameras with wide angle lens (e.g., fisheye and catadioptric)

The parameters in these files are typical values. If you wish to change the parameters, please refer to the comments in these two files.

Camera calibration files

If you want to use your own camera, make sure a global shutter camera is used. A good choice is the Bluefox camera from MatrixVision. You can put camera calibration files under svo_ros/calib and load them as in run_from_bag.launch. We use yaml files to specify camera parameters. Please refer to calibration.md for more details.

Inertial-aided frontend

Without using the relatively heavy ceres-based backend, the frontend can also use IMU to facilitate visual tracking. This is not as accurate as a tightly coupled VIO, but is relatively lightweight.

We provide two launch files for EuRoC for monocular and stereo setups:

roslaunch svo_ros euroc_mono_frontend_imu.launch
roslaunch svo_ros euroc_stereo_frontend_imu.launch

These launch files read the parameters from param/euroc_mono_imu.yaml and param/euroc_stereo_imu.yaml. The parameters are not necessarily optimal for every sequence, but should be enough as a good starting point.

The first several images of many EuRoC sequences are not good for initialize SVO, especially for the monocular case. Therefore it is better to start the bag at a later time, for example:

rosbag play MH_01_easy.bag -s 50
rosbag play MH_02_easy.bag -s 45
rosbag play V1_01_easy.bag
rosbag play V1_02_medium.bag -s 13
rosbag play V2_01_easy.bag
rosbag play V2_02_medium.bag -s 13

This is to avoid, for example, strong rotation for monocular initialization. For more details on using SVO with this configuration, please read the step-by-step instruction.