vilib_ros

ROS wrapper for CUDA Visual Library by RPG.

Tested with ROS melodic with custom OpenCV installation. (3.4.10)

Dependencies

The following packages are required for the vilib_ros wrapper.

CUDA

Ensure that the CUDA toolkit has been installed: https://developer.nvidia.com/cuda-toolkit

vilib

The library could be found here, follow the installation steps to install the library.

ROS Packages

• vision_opencv: $ sudo apt install ros-melodic-vision-opencv
• image pipeline: $ sudo apt install ros-melodic-image-pipeline

Nodes

There are 2 nodes for this package, which are:

• vilib_ros_node: FAST corner detection via the vilib FAST library
• vilib_ros_tracker: Lucas - Kanade Feature Tracking via the vilib FAST library

FAST Corner Detection

vilib_ros_node

/img_in (sensor_msgs::Image) [Subscription]

Input image from camera.

/fast_pts (vilib_ros::keypt) [Publication]

FAST feature keypoints detected from the image.

/img_out(sensor_msgs::Image) [Publication]

Image overlayed with detected keypoints.

Feature Tracking (Lucas - Kanade Feature Tracking)

vilib_ros_tracker

/img_in (sensor_msgs::Image) [Subscription]

Input image from camera.

/feature_pts (vilib_ros::keypt) [Publication]

Track points detected from the image.

/img_out(sensor_msgs::Image) [Publication]

Image overlayed with detected tracking points.

Custom message

vilib_ros::keypt

A custom msg type is used to publish the various feature keypoints in the /fast_pts topic. The structure is as follows:

time stamp: The timestamp when the topic is published.
geometry_msgs/Point[] points: The FAST features detected, stored as an array of geometry_msgs/Point.
int32 size: The number of points detcted in the current frame.

Note: The z parameter of the geometry_msgs/Point in vilib_ros_node is not used, whereas the z parameter of the geometry_msgs/Point in vilib_ros_tracker is used as the tracking point ID.

Dynamic Reconfigure

This wrapper supports dynamic reconfiguration of the various parameters for the nodes! Launch rqt_reconfigure and adjust the parameters accordingly.

$ rosrun rqt_reconfigure rqt_reconfigure

FAST Corner Detection

Frame preprocessing

• PYRAMID_LEVELS: Pyramid level

FAST detector parameters

• FAST_EPSILON: Threshold for feature detection
• FAST_MIN_ARC_LENGTH: Threshold for feature detection
• FAST_SCORE: FAST_SCORE Setter (Sum of absolute differences/Maximum threshold value)

Non-Maximum Suppression (NMS) parameters

• HORIZONTAL_BORDER: Horizontal image detection padding
• VERTICAL_BORDER: Vertical image detection padding
• CELL_SIZE_WIDTH: NMS cell width
• CELL_SIZE_HEIGHT: NMS cell height

Feature Tracking

Frame preprocessing

• FRAME_IMAGE_PYRAMID_LEVELS: Image pre-processing pyramid level

Feature Detection parameters

• FEATURE_DETECTOR_CELL_SIZE_WIDTH: NMS cell width
• FEATURE_DETECTOR_CELL_SIZE_HEIGHT{32}: NMS cell height
• FEATURE_DETECTOR_MIN_LEVEL: Image pre-processing pyramid level (Minimum)
• FEATURE_DETECTOR_MAX_LEVEL: Image pre-processing pyramid level (Maximum)
• FEATURE_DETECTOR_HORIZONTAL_BORDER: Horizontal image detection padding
• FEATURE_DETECTOR_VERTICAL_BORDER: Vertical image detection padding

FAST Parameters

• FEATURE_DETECTOR_FAST_EPISLON: Threshold for feature detection
• FEATURE_DETECTOR_FAST_ARC_LENGTH: FAST arc length
• FEATURE_DETECTOR_FAST_SCORE: FAST_SCORE Setter (Sum of absolute differences/Maximum threshold value)

Troubleshooting

• If there is compilaion issue of cv_bridge caused by the being unable to locate the OpenCV path, the cv_bridge cmake have to be updated to point to the new path. For example, updating the path

if(NOT "include;/usr/include;/usr/include/opencv" STREQUAL " ")

  set(cv_bridge_INCLUDE_DIRS "")
  set(_include_dirs "include;/usr/include;/usr/include/opencv")

to

if(NOT "include;/usr/local;/usr/local/opencv/include" STREQUAL " ")

  set(cv_bridge_INCLUDE_DIRS "")
  set(_include_dirs "include;/usr/local/opencv/include;/usr/local/opencv/include/opencv2")

License

Licensed under the MIT License.