Uporaba robotskega vida za določitev trajektorije leta objekta v realnem času

Hauptman, Anamarija (author), Mihelj, Matjaž (mentor), Gams, Andrej (co-mentor)

V sklopu diplomskega dela smo razvili algoritem, ki v realnem času sledi žogico med njenim preletom vidnega polja kamere. Poleg tega algoritem določi tudi trajektorijo leta žogice. Najprej smo se spopadli z izzivom kalibracije kamere. Poleg notranje kalibracije kamere, ki je nujna za razrešitev problema distorzije lece, smo morali izvesti tudi zunanjo kalibracijo, s katero smo dobili pretvorbo med slikovnimi elementi na sliki in milimetri v realnem svetu. Nato smo morali ugotoviti, kako bi na sliki sploh lahko prepoznali žogico. Preizkusili smo tri metode: barvni filter, zaznavanje gibanja in Houghova transformacija ter različne kombinacije teh treh metod. Za najučinkovitejšo se je izkazala funkcija, ki temelji na metodi barvnega filtra. V zadnjem delu diplome smo določili matematični model leta žogice. Z metodo vsote najmanjših kvadratov smo na podlagi točk, kjer je algoritem zaznal žogico na preteklih sličicah, določili trajektorijo, po kateri se je žogica gibala.

• Language: Slovenian
• Keywords: računalniški vid, kalibracija kamere, OpenCV, vsota najmanjših kvadratov
• Work type: Bachelor thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2023
• PID: 20.500.12556/RUL-148799
• COBISS.SI-ID: 163035907
• Publication date in RUL: 31.08.2023

Secondary language

• Language: English
• Title: Robot vision application to determine flight path of an object in real time

Abstract

As part of the thesis, we developed an algorithm that tracks the ball in real time as it flies through the field of view of the camera. In addition, the algorithm also determines the trajectory of the ball’s flight. First, we tackled the camera calibration challenge. In addition to the internal calibration of the camera, which is necessary to solve the problem of lens distortion, we also had to perform an external calibration, with which we obtained the conversion between the pixels in the image and millimeters in the real world. Then we had to figure out how to identify the ball in the picture. We tested three methods: color filter, motion detection and Hough transform, as well as various combinations of these three methods. The function based on the color filter method proved to be the most effective. In the last part of the thesis, we determined the mathematical model of the ball’s flight. Based on the point where the algorithm detected the ball in the previous frames, we determined the trajectory along which the ball moved using the least squares method.

• Keywords: computer vision, camera calibration, OpenCV, least square method