Automatic pattern detection on images for multiple-sensor calibration

Abstract

Sensor calibration can be done by identifying known points of interest of a calibration pattern into the intensity images captured with the sensor. If the same exact points are identified using multiple sensors, then multiple-sensor calibration can be achieved and the results used for sensor fusion. In this master thesis, several scripts for automatic detection of intersections of a calibration pattern imaged with three types of sensors: an RGB camera, a thermal camera, and a lidar system have been developed. The aim has been to pursue an automatic way for extracting the points of interest employing digital image processing techniques. Three different image sets of a calibration pattern, one set for each sensor, have been used as input data and the intersections obtained as results. In the case of RGB images, fully automatic operation (without user interaction) has been achieved, detecting the region of interest (ROI) and extracting the intersection points. For thermal and lidar images, the user is required to select the ROI before the script can find the intersection points.