Movement andcontrol of mobilerobots inclosed workspacesdepends on globalknowledge of theenvironment and exactknowledge of robotposition atalltimes(localization). Odometry, which is a popular relative localization method, is prone to systemic and non-systemicerrors andtherefore, results inthe inaccurate estimation ofthe position of therobot. Consequently, localization systemsthat relyon sensingsystems like visioncameras, rangearegaining importance inthedomain. This project deals withimplementingone suchsystemthat usesinfrared cameras tocapture data ofthe position ofseveral robots inapre-defined workspace.This is an absolute localization for severalrobots equipped with IR LEDs, using multiple infrared cameras.
In the workspace of the Multi-Robot Localization,system has four cameras mounted on the ceiling of the project room and they are placed on the 4 corners of a (≈1x1m2) square. Each camera has a triangular pyramid detection range. These ranges are not independent; they have overlapping space in which the robots (LEDs) can be detected by more than one camera at the same time. The base surface of each camera’s detecting range is a rectangle.
The 4 cameras in the localization system sends the data via CAN bus. ROS drivers were developed to read these CAN bus data and convert them to camera cordinates. Then the simulation environment is set up in rviz which displays the field of view of the camera as well as the detected points. Since the main objective was to make the system modular and scale up when required, A calibration method needs to be introduced for the efficient calibration of the cameras when scaling up the array. ROS nodes were created to calibrate these cameras and special calibratin boards made up of IR leds were made. An accuracy of 5 cm were achieved in localization of robots after the calibraion.