Show EOL distros: 

Package Summary

This package contains a simulated object recognizer which published objects based on preset poses and the current camera frames

Package Summary

This package provides a 'perception algorithm'-independent simulation of 6-D object localization for 3D object search by a mobile robot: Based on the poses of the searched objects with respect to the current viewing frustum(s) of the robot, the detectability of the objects is estimated.

Package Summary

This package provides a 'perception algorithm'-independent simulation of 6-D object localization for 3D object search by a mobile robot: Based on the poses of the searched objects with respect to the current viewing frustum(s) of the robot, the detectability of the objects is estimated.

Logo-white-medium.jpg

Description

This package provides a 'perception algorithm'-independent simulation of 6-D object localization for 3D object search by a mobile robot: Based on the poses of the searched objects with respect to the current viewing frustum(s) of the robot, the detectability of the objects is estimated.

Functionality

The system loads objects specified in a object-constellation-file and recognizes them based on the current camera position (for both, left and right camera). For a single object at first the given pose is transformed from the world to the camera frames. For each camera frame the transformed pose is rated based on its position in the camera frustum (the distance to the camera and the angles to the camera's visual axis (azimut and elevation) are used). If the rating is higher than a specified threshold value the pose is valid and the object is marked as found and returned.


Optionally random errors which can occur in real world recognition scenarios can be simulated in three ways:

  1. To simulate frame specific recognition errors random correctly found objects can be dropped every couple frames
  2. To simulate the recognition accuracy in the found position of the object poses random noise is added to slightly alter the position in all dimensions
  3. Similar to the position noise errors in the orientation can be generated for each dimension


The object-constellation file is using the XML-format and needs to have a specific layout:

  • The file needs to have at least a root node called Objects

  • All recognizable objects are child nodes of this root node and are named Object

  • The value of each object is its pose (position in x-, y-, z-direction followed by its orientation as a quaternion (w,x,y,z) or euler angles)
  • In addition to the pose each object has 4 attributes (similar to the meta information of the objects in the asr_object_database):

    • type: The name of the object
    • mesh: The path to the object's mesh file (as a package-url)
    • id: The id of the object
    • angles: Specifies how the orientation is denoted in the pose (can be either quaternion or euler)

An example for a valid constellation file could be the following:

<Objects>
    <Object type="Cup" id="011021054100" mesh="package://asr_object_database/rsc/databases/segmentable_objects/Cup/object.dae" angles="quaternion">-1.2902,0.729374,0.755761,0.721985,-0.665815,-0.13442,0.131754 </Object>
    <Object type="CoffeeBox" id="0" mesh="package://asr_object_database/rsc/databases/textured_objects/CoffeeBox/CoffeeBox.dae" angles="euler">1.5,-0.5,1.5,90,0,0 </Object>
</Objects>

Usage

Needed packages


If the feature is not turned off there also needs to be a valid transformation between the frames of the camera and the ones of the objects in the configuration file (by default \map). For more information on the transformation check asr_kinematic_chain_dome, asr_kinematic_chain_mild and tf.

Needed software

Needed hardware

No hardware is needed for this package to work but by default there has to be a camera frame which can either be simulated or gathered by using a real camera setup.

Start system

At first set the configuration parameters depending on your usage scenario (see configuration). After you have chosen the object constellation you want to use and set the topics correctly start the process:

roslaunch fake_object_recognition fake_object_recognition.launch


Now call one of the offered services to add or remove objects from the list of detectable entities.

ROS Nodes

Subscribed Topics

Apart from the frames mentioned above there are no input topics for this package.

Published Topics

  • visualization_msgs/Marker: A visualization of all available objects in the chosen constellation (default: /asr_fake_object_recognition/constellation_visualization)

  • asr_msgs/AsrObject: The found objects as an AsrObject -message to be used by other packages which depend on the results of object recognizers (default: /stereo/objects)

  • visualization_msgs/Marker: A visualization of the found objects depending on the current camera pose (default: /stereo/visualization_marker)

Parameters

This package offers two sets of parameters you can adjust, the static ones which you can set by adjusting the params.yaml file located in the param-directory of the package, and the dynamic ones which you can either set by adjusting the launch-file or during runtime by using dynamic_reconfigure.

Static parameters:

  • frame_world (string): The name of the frame the objects in your constellation file are relative to

  • frame_camera_left (string): The name of the left camera frame in your (simulated or real) camera setup

  • frame_camera_right (string): The name of the right camera frame (Should be the same as frame_camera_left if you are using a mono camera setup)

  • config_file_path (string): The path to the XML-file containing your object constellation (the path can be relative to your package root or absolute)

  • output_rec_objects_topic (string): The name of the topic the found objects are published on as AsrObject -messages

  • output_rec_marker_topic (string): The name of the topic the visualization of the found objects is published on

  • output_constellation_topic (string): The name of the topic the visualization of all available objects in your constellation-XML is published on

  • rating_theshold (double): The threshold value for the distance rating (objects are found if their rating is higher than this value)

  • rating_threshold_x (double): The threshold value for the azimut angle rating (objects are found if their rating is higher than this value)

  • rating threshold_y (double): The threshold value for the elevation angle rating (objects are found if their rating is higher than this value)

  • timer_duration (double): The time between the recognition cycles (in seconds)

Dynamic parameters:

  • frustum_mode (integer): Specifies which camera frustum is used for the pose rating of the object (0=both needed, 1=only one needed, 2=use the left one, 3=use the right one)

  • use_camera_pose (boolean): If false all objects in the constellation-config-file are published without checking their pose relative to the camera

  • use_pose_invalidation (boolean): If true randomly invalidates found poses based on prob_pose_invalidation-value

  • prob_pose_invalidation (double): The probability which is used to invalidate poses

  • use_position_noise (boolean): If true randomly alters the positions of the found objects in all dimensions depending on a normal distribution

  • pos_noise_normal_dist_mean (double): The mean value of the normal distribution used for the generation of position errors

  • pos_noise_normal_dist_dev (double): The standard deviation value of the normal distribution used for the generation of position errors

  • use_orientation_noise (boolean): If true randomly alters the orientations of the found objects depending on normal distributions

  • or_x_noise_normal_dist_mean (double): The mean value of the normal distribution used for the generation of orientation errors (x-axis)

  • or_x_noise_normal_dist_dev (double): The standard deviation value of the normal distribution used for the generation of orientation errors (x-axis)

  • or_y_noise_normal_dist_mean (double): The mean value of the normal distribution used for the generation of orientation errors (y-axis)

  • or_y_noise_normal_dist_dev (double): The standard deviation value of the normal distribution used for the generation of orientation errors (y-axis)

  • or_z_noise_normal_dist_mean (double): The mean value of the normal distribution used for the generation of orientation errors (z-axis)

  • or_z_noise_normal_dist_dev (double): The standard deviation value of the normal distribution used for the generation of orientation errors (z-axis)

  • config_file_path (string): The path to the XML-file containing your object constellation (the path can be relative to your package root or absolute; at startup the path set in the config.yaml file is used and can be altered with this parameter during runtime)

Needed Services

No external services are called by this package.

Provided Services

  • get_all_recognizers: Adds all objects in the used constellation file to the list of recognizable objects

    • Parameters: -
    • Returns: -
  • get_recognizer: Adds a single object in the used constellation file to the list of recognizable objects

    • Parameters:
      • object_type_name (string): The name of the object to add
    • Returns: -
  • clear_all_recognizers: Clears the list of recognizable objects

    • Parameters: -
    • Returns: -
  • release_recognizer: Removes a single object from the list of recognizable objects

    • Parameters:
      • object_type_name (string): The name of the object to remove
    • Returns: -

Wiki: asr_fake_object_recognition (last edited 2020-02-16 16:17:09 by PascalMeißner)