Computer Vision Development

Overview

Computer Vision plays an import role in this competition. Whether you can win a match is directly depends on whether you can shoot on the target (opponent robots’ armor). In addition, Competition Committee has added April Tag for fiducial navigation, Sentinel Cameras for visual localization opponent’s robots and different armor hitting points for encouraging pose recognition. It is fair to say that the Computer Vision contains at least 50% of the potential topics in this challenge.

Platform

Real Robot

  • Input: Daheng Mercure Camera as first person view of the robot
  • Secondary Input: USB camera as the sentinel from corners
rplidar_a3

Input data

  • Pictures from both cameras
  • Pictures from VCATS

Environment

  • The codes can work in any OS system and with any programming language, but it needs to be convertible to Ubuntu 18.04 (arm64) and C++

Targets

  • The whole robot and its pose
  • The robot armor board
  • The light bar on the back of the robot
  • The April Tag on the walls

Armor detection

  • The target is the lighted (blue or red) armor board on the body of the robot.
  • The objective is to select the exact area (by pixel) in the picture.
  • A very important constraint is the time limit. Each robot can move up to 2 meters per second. A 0.1 second delay results in 0.2 meter deviation and misses the target (armor board is 0.15*0.15).
  • At the center of armor board, each robot from both team has a sticker showing “1” or “2” as identification for the robot. It would be better to identify which robot we are aiming to. 
  • The potential research project will be to try different object detection algorithms and design experiments for comparison.

Pose Recognition

  • The aim is to detect which armor board is detected by armor detection algorithm. However, since each armor board is identical on one robot, an obvious choice would be to detect opponents robot pose.
  • There is no current solution on this problem, so that any potential solution is welcomed.
  • The potential research project will be to try different pose recognition algorithms and design experiments for comparison.

fiducial navigation

  • The target pictures are in April Tag.
  • It is recommended to use traditional image processing method to do this as the target images are limited.
  • The aim for this would be to use robot front camera to move to specific location using April Tag detection as fiducial navigation system.
  • The potential research project will be to try different algorithms and design experiments for comparison.

Visual Localization

  • The Sentinel Cameras are installed on the diagonal corners about 1.5-2 meters above the floor.
  • The Sentinel Cameras have a separate computing device that can keep communicating with both our robots.
  • The key information that provided by Sentinel Cameras should be the location of opponents, especially when our robots front cameras were blocked by walls.
  • Object Tracking could be a potential solution for this problem.
  • The potential research project will be to try different algorithms and design experiments for comparison.

If interested, what's next?

If you are interested to choose one of the above projects are your capstone project, please click this link. Otherwise, please read the following points:

  1. Most of these project is recommended to develop and explore as a Team since the scale for each topic is pretty big. Please fill in the form that we sent by email after induction section, so that it would be easy for us to manage different teams with different projects.
  2. All projects related with Computer Vision are expected to be converted in a Ubuntu OS (arm64 frame) with C++, there are some instruction for install ROS available on this link.
  3. Please contact Guang (ghu1@student.unimelb.edu.au) by email if you have any question about Computer Vision Projects.