Identifying Pokémon Cards
summary
A tool digitizes a Pokémon TCG inventory by grabbing webcam frames, extracting a card with edge detection and a contour->line->corner pipeline, warping to a 2D scan and matching against a scraped 14,000 card dataset. Implementation uses grayscale + Sobel, a voting “half” transform for edges and n-algebra for the perspective matrix, then a 16x16 perceptual hash (256 bits) for fast lookup and per-symbol template matching for set icons. Works well on a white background; foils and glare still break matching.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:08Overview
- 1:32Background Noise
- 2:01Data Set
- 2:26Detection
- 3:46Image Filter
- 5:21Contour
- 8:10Perspective
- 8:50Searching for matches
- 9:11Perceptual hashes
- 9:56Binarization
- 10:35Full process
- 11:04The problem
- 11:37Template matching
- 12:19Testing
- 13:22Outro
related talks
48mHow I found my Pokémon cards thanks to Postgres: an AI journey
26mIs This Your Card? Computer Vision for Playing Card Recognition
31mBecoming a Pokémon Master with DVC: reproducible machine learning experiments
30mGenerating embeddings for Yu-Gi-Oh Cards with NumPy
34mCard detection using YOLO on Android
14mComputer vision on edge devices
30mIntroduction to Computer Vision with OpenCV
22mHow to effectively extract image features – let’s play with...