Razpoznavanje šahovskih figur iz 3D oblakov točk za potrebe robotskega šaha

BORŠTAR, VILIJEM (author), Perš, Janez (mentor), Grm, Klemen (comentor)

Avtonomna igra šaha proti človeku je izziv, ki ima dolgo zgodovino s preboji, ki v današnjem času igrajo pomembno vlogo na različnih področjih. V zgodovini je človeštvu uspelo osvojiti šahovsko logiko in izziv premikanja figur po šahovnici, medtem ko je naloga razpoznave šahovskih figur izziv, ki še danes ni v celoti rešen. Namen našega magistrskega dela je raziskati možnost uporabe kamere Intel RealSense D435 in modula OpenCV Surface Matching, ki temelji na uporabi značilk Point-Pair za razpoznavo šahovskih figur. Obenem smo raziskali vpliv dejavnikov osvetlitve, odsevnosti, zasuka in prekrivanja figur, ter gostote zajetega oblaka točk na uspešnost razpoznave figur. V ta namen smo izdelali svoje 3D-modele figur, jih natisnili in izdelali nabor oblakov točk, kjer so bile natisnjene figure slikane pri različnih pogojih. Na temelju rezultatov smo prišli do zaključka, da razpoznava šahovskih figur s kamero Intel RealSense D435 in modulom OpenCV Surface Matching ni dovolj zanesljiva za igro šaha proti človeku, vendar smo vseeno dobili boljšo predstavo o vplivu omenjenih dejavnikov na uspešnost razpoznave šahovskih figur, kar bo uporabno pri nadaljnjem raziskovanju razpoznave predmetov iz oblakov točk.

• Language: Slovenian
• Keywords: računalniški vid, Surface Matching, značilke Point-Pair, Intel RealSense D435, šah
• Work type: Master's thesis/paper
• Organization: FE - Faculty of Electrical Engineering
• Year: 2023
• PID: 20.500.12556/RUL-145320
• COBISS.SI-ID: 149513731
• Publication date in RUL: 17.04.2023

Secondary language

• Language: English
• Title: Recognizing Chess Pieces From 3D Point Clouds for the Purpose of Robotic Chess

Abstract

The autonomous game of chess against a human opponent is a challenge with a long history of breakthroughs that play an essential role in various fields today. Throughout history, humanity has managed to master chess logic and the challenge of moving chess pieces around the chessboard, while the task of recognizing chess pieces is a challenge that has not been fully solved even today. This paper aims to research the possibility of using an Intel RealSense D435 camera and the OpenCV Surface Matching module, which is based on using Point-Pair features to recognize chess pieces. Additionally, we investigated the influence of several factors, namely illumination, reflectivity, rotation and overlapping of pieces, and the density of the captured point cloud on the success of chess piece recognition. For this purpose, we designed 3D models of chess pieces, printed them and created a set of point clouds, where the printed pieces were imaged under different conditions. Based on the results, we concluded that chess piece recognition with the Intel RealSense D435 camera and the OpenCV Surface Matching module is too unreliable to be used in an automated chess game against a human opponent. However, it has given us a better understanding of the influence of the mentioned factors on the success of chess piece recognition, which may prove helpful in further research on object recognition from point clouds.

• Keywords: Computer Vision, Surface Matching, Point-Pair Features, Intel RealSense D435, Chess