A Monocular Vision-based System Using Markers for a Real-Time 6D Pose Estimation of a Trailer
Author
Mark, Natalie
Term
4. semester
Education
Publication year
2022
Pages
56
Abstract
Dette projekt undersøger objektdetektion og poseestimering med et enkelt, almindeligt (monokulært) kamera uden brug af neurale netværk. Målet er en trailer. I stedet for at genkende traileren direkte er der designet en simpel visuel markør, som kan monteres på den: to farvede kugler, en blå og en orange. Systemet finder markøren med klassiske computervisionsmetoder—farvesegmentering for at isolere farverne og Hough-cirkeldetektion for at lokalisere kuglerne i billedet—og estimerer derefter trailerens pose (dens position og orientering) ud fra kuglernes indbyrdes placering. Rapporten gennemgår også eksisterende markørdesign og detektionsmetoder samt sammenligner forskellige sensorsystemer med deres vigtigste fordele og ulemper. I test virker poseestimeringen på afstande op til 560 mm, forudsat at begge kugler er i kameraets synsfelt. Afslutningsvis diskuteres både forventede og uventede resultater samt mulige spor for fremtidigt arbejde.
This project explores object detection and pose estimation using a single, standard (monocular) camera without neural networks. The target object is a trailer. Instead of recognizing the trailer directly, a simple visual marker is designed to be attached to it: two colored spheres, one blue and one orange. The system detects the marker with classical computer-vision techniques—color segmentation to isolate the colors and Hough circle detection to locate the spheres in the image—and then estimates the trailer’s pose (its position and orientation) from the spheres’ relative locations. The report also reviews existing marker designs and detection methods, and compares different sensor systems, outlining their main advantages and disadvantages. In testing, pose estimation works at distances up to 560 mm, provided both spheres are within the camera’s field of view. The report concludes with a discussion of expected and unexpected results and directions for future work.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Keywords
Documents