Download presentation
Presentation is loading. Please wait.
Published byWarren Howard Modified over 9 years ago
1
Deep Green System for real-time tracking and playing the board game Reversi Nadav Erell Intro to Computational and Biological Vision, CS department, Ben-Gurion University 22 March, 2013
2
Reversi Nadav Erell, Intro to Computational and Biological Vision 2013
3
Game Rules Nadav Erell, Intro to Computational and Biological Vision 2013
4
Game Rules Nadav Erell, Intro to Computational and Biological Vision 2013
5
System Overview Capture Frame o Standard webcam o Simple setup o Flexible Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI Capture Frame Detect Board area o Segmentation by colors o Corner detection o Homography Capture Frame Detect Board area Detect Discs Send acquired board state to game AI o Game representation o Technique for AI
6
System Overview Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
7
System Overview Classic segmentation problem Best clue in the image for identifying the board? It’s green! Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI It’s a start, but really not good enough…
8
System Overview Solution: Thresholding on each channel Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
9
System Overview Solution: Thresholding on each channel Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
10
System Overview Solution: Thresholding on each channel Combining the masks Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI Combined Mask After Dilation and Erosion
11
System Overview Solution: Thresholding on each channel Combining the masks Corner identification Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
12
System Overview Homography Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
13
System Overview Discs from background: Red channel thresholding Filtering connected components o Total area o Eccentricity Matching remaining discs to grid Separating white and orange discs using blue channel Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
14
System Overview Perfect Detection! Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
15
Artificial Intelligence Final detected board state is represented as an 8x8 matrix Game is fully independent of the past – can analyze only current state without saving data Game tree search o Minimax algorithm o Pruning Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
16
Artificial Intelligence Suggested moves by AI: Nadav Erell, Intro to Computational and Biological Vision 2013 Capture Frame Detect Board Area Detect DiscsReversi AI
17
Highlights Reversi is a fun game. Give it a try if you never played! System is easy to set up, simple hardware Robust and flexible in camera angle and distance Acceptable latency for real-time game tracking Nadav Erell, Intro to Computational and Biological Vision 2013
18
Questions?
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.