Frank Engelhardt's profile - activity

Is it maybe sufficient, if you specify the desired timestamp in the message headers, rather than the actual generation time (I guess you would use ros::Time::now())?

For B, synchronizing with the desired timestamp is easier and avoids using message_filters/approximate_time_policy. There is always an unknown delay for B for message reception due to transmission times an OS scheduling, so generating messages at exact time instants does probably not give you good results either.

When I understand you right, the features can be filtered out by their depth value. I wonder if my features disappear simply because their depth may be hard to determine. The disparity map contains many undefined pixels...

My baseline according to right/camera_info is 0.4m, which matches my robot's configuration. The building is about 30m long. Settling Odom/MaxDepth to 0 does not do the trick.

Exit code -6 might be an assertion error. Maybe you should post the screen outputs and log file contents.

I am using rtabmap_ros with a simulated stereo camera in gazebo, but I am struggeling to get the stereo_odometry node computing any transformation.

The problem is that too few (mostly not even one) features are found in my camera images. Which is confusing me, since my own implementation of GFTT finds plenty of features in the same images.

Am I getting anything wrong? Are there more requirements on the visual features for rtabmap for visual odometry?

Considering the following setup, where I am using GFTT feature detector:

<node pkg="rtabmap_ros" type="stereo_odometry" name="stereo_odometry">
    <remap from="left/image_rect"       to="/stereo_camera/left/image_rect"/>
    <remap from="right/image_rect"      to="/stereo_camera/right/image_rect"/>
    <remap from="left/camera_info"      to="/stereo_camera/left/camera_info"/>
    <remap from="right/camera_info"     to="/stereo_camera/right/camera_info"/>
    <remap from="odom"                  to="/stereo_odometer/odometry"/>

    <param name="frame_id" type="string" value="base_link"/>
    <param name="odom_frame_id" type="string" value="odom"/>
    <param name="publish_tf" value="true" />

    <param name="Odom/InlierDistance" type="string" value="20"/>
    <param name="Odom/MinInliers" type="string" value="10"/>

    <param name="Odom/FeatureType" type="string" value="4"/>
    <param name="GFTT/UseHarrisDetector" type="string" value="false"/>
    <param name="GFTT/BlockSize" type="string" value="3"/>
    <param name="GFTT/K" type="string" value="0.1"/>
    <param name="GFTT/MinDistance" type="string" value="20"/>
    <param name="GFTT/QualityLevel" type="string" value="0.005"/>
    <param name="Odom/MaxFeatures" type="string" value="1000"/>
    <param name="Odom/Strategy" type="string" value="1"/>
</node>

The output of rostopic echo /odom_info says, I get a total amount of 2 recognized features in a particular scene (see attached image below):

---
header: 
  seq: 387
  stamp: 
    secs: 1615
    nsecs: 663000000
  frame_id: odom
lost: True
matches: 2
inliers: 0
variance: 0.0
features: 0
localMapSize: -1
timeEstimation: 0.00351881980896
timeParticleFiltering: 4.57832234264e-41
stamp: 1615.66296387
interval: 0.10000000149
distanceTravelled: 0.0
type: 1
wordsKeys: []
wordsValues: []
wordMatches: []
wordInliers: []
refCorners: 
  - 
    x: 599.0
    y: 419.0
  - 
    x: 602.0
    y: 421.0
newCorners: 
  - 
    x: 323.645477295
    y: 419.167877197
  - 
    x: 332.146850586
    y: 421.871734619
cornerInliers: []
transform: 
  translation: 
    x: 0.0
    y: 0.0
    z: 0.0
  rotation: 
    x: 0.0
    y: 0.0
    z: 0.0
    w: 0.0
transformFiltered: 
  translation: 
    x: 0.0
    y: 0.0
    z: 0.0
  rotation: 
    x: 0.0
    y: 0.0
    z: 0.0
    w: 0.0
---

However, my own implementation of GFTT in OpenCV using the same parameters finds a plenty of features. The following image shows the features from my implementation. The same image was fed into the stereo_odometry node. Is there any hint why stereo_odometry finds only two features?

Edit: Example output of stereo_odometry

[ WARN] (2015-09-30 11:23:52.840) util3d_registration.cpp:173::transformFromXYZCorrespondences() RANSAC refineModel: Refinement failed: got an empty set of inliers!
[ WARN] (2015-09-30 11:23:52.840) OdometryOpticalFlow.cpp:403::computeTransform() Transform not valid (inliers = 0/14)
[ INFO] [1443605032.840375078, 364.936000000]: Odom: quality=0, std dev=0.000000m, update time=0.005786s

Edit2 : Here is an example disparity map of my scene: https://goo.gl/photos/6SPKukujRP6fBQpLA

Edit3 : example bag file: https://drive.google.com/file/d/0B_p8...

Using libpcl-1.7.1 works for me.

I have downloaded and installed it from source: https://github.com/PointCloudLibrary/...

In rgbdslam's CMakeLists.txt change

find_package(PCL 1.7 REQUIRED COMPONENTS common io)

to

find_package(PCL 1.7.1 EXACT REQUIRED COMPONENTS common io)