Measurements of Geometric Transformations in Anechoic Chambers using Computer Vision

Abstract

Abstract The process of developing an antenna system requires several important steps with simulations and measurements. The test phase is to verify that the device fulfills the specifications and performs as expected from the simulations. The device is tested in an anechoic chamber, which is a controlled environment with radiation-absorbent material, where it is possible to achieve good measurements with very little noise and disturbances. The testing itself is a long process and can take several days or even weeks to finish. Before a test the device is mounted on a movable rig and the device has to be aligned in a specific way to avoid pointing errors in the radiation pattern. This alignment is done manually with a spirit level and a laser. The aim of this thesis is to develop a method that could automate the alignment process and the goal is to achieve an accuracy comparable with spirit level and laser. The basis for the thesis was computer vision, which was used to analyse images of three devices from two anechoic chambers. The devices had in total four degrees of freedom, one translation and three rotations, yaw, pitch and roll. The used method was based on detecting corners of the devices in the images, where Shi-Tomasi corner detection was used and compared with a manual corner detection. Both techniques were combined with bundle adjustment. The best result shows that it is possible to achieve a mean offset of 0.25 degrees in the yaw rotation compared to the traditional method with a spirit level and laser. The pitch rotation gave a mean offset of 0.96 degrees, the roll rotation gave a mean offset of 0.24 degrees and the translation direction gave a mean offset of 0.35 centimeter. Based on the results it is possible to automate the alignment process with computer vision and achieve good accuracy for small angles.