Visual tracker package

---

Visual feature tracker for visual servoing and targeting

This package contains 2 feature trackers to be used for feature targeting or visual servoing. It also contains a graphical interface for the user to select the desired target and send messages to the autonomous program.

REQUIREMENTS

• Python 2
• ROS Melodic
• OpenCV > 3.4
• cv_bridge
• GUI exclusive:
  • Pillow (python)
  • Tkinter (python)
  • image_transport for image decompression

Usage

Run one of the tracker nodes (csrt_tracker.py or lt_tracker.py). These subscribe to an image topic, e.g. /usb_cam/image_raw, the user selected target to track /gui_target and the current position to the wall, e.g. /laser_wall/wall_position.

The CSRT tracker, which is the latest calculates the target position and calculates the transformations from target to drone and to the wall so all the frames can be view in RViz. If the position of the camera is changed this should also be changed in the script so the correct drone position is calculated. The tracker should be run on the drone.

The GUI is launched with the launch-file gui.launch which also starts the image decompression node. The GUI shows the currently decompressed image on /camera/image_decompressed. Using the GUI the user can get current status information as well as use the buttons to change the flight behavior and select a target to track. Run the GUI on the users computer and not on the drone mounted computer.

Scripts

csrt_tracker.py

This tracker uses the CSRT tracker implemented in OpenCV. When a point is selected a 30x30 pixel bounding box is selected around the point and tracked. If the bounding box estimate is to large or small it reinitializes a new bounding box. There is minimal error / loss of target as it is only run in a supervised setup. On the NUC it runs at approximately 20Hz.

lt_tracker.py

Tracker based on the bachelor project of Laura and Tobias. It tracks and matches ORB features between two consecuitive images. This tracking scheme assumes that all the features are on the same plane right in front of the camera so that they will all move in a similar manner. From this assumption the feature selected will also have moved the same distance. The topics setup of this script is not up to date with the rest of the setup, but the tracker itself works.

visual_gui_v2.py

GUI for selecting the target and starting autonomous flight for the controller.

visual_gui.py (OLD)

Old version of the gui. Similar to v2 and Might work, but has less features.