Skip to content

Latest commit

 

History

History
 
 

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

groundHOG

This package uses cuda to extract HOG features and detect persons just using the RGB image of the Kinect.

Dependencies

  • libcudaHOG: see rwth_preception_people/3rd_party If this dependency is not met, the package will still compile but won't have any functionality. This is meant to not break the overall build process on systems without a NVIDIA graphics card.

Run

Parameters:

  • queue_size default = 20: The synchronisation queue size
  • model default = "": The model file, can be found in model/config
  • image_color default = /camera/rgb/image_color: The Kincet colour image
  • camera_info default = /camera/rgb/camera_info: The Kinect camera info
  • ground_plane default = "": The ground plane. Published by upper_body_detector. Will only be used if set. Is used to track speed up detections.
  • detections default = /groundHOG/detections: The generated data output topic
  • result_image default = /groundHOG/image: The generated image output topic showing the detections as boundingboxes.

rosrun:

rosrun rwth_ground_hog groundHOG [_parameter_name:=value]

roslaunch:

  • Using only the RGB image: roslaunch rwth_ground_hog ground_hog.launch [parameter_name:=value]
  • Using ground plane to enhance detection roslaunch rwth_ground_hog ground_hog_with_GP.launch [parameter_name:=value]

Credits

Original author: Patrick Sudowe, Computer Vision Group, RWTH Aachen
ROS integration: Christian Dondrup (@cdondrup), LCAS, University of Lincoln
SPENCER integration: Stefan Breuers, Computer Vision Group, RWTH Aachen