Interactive 'Higher or Lower' Game using Computer Vision

💡 Overview

This project explores the intersection of Human-Computer Interaction (HCI) and game development. Instead of using a mouse or keyboard, players interact with the game using physical playing cards. The system uses a camera to "see" the card, identify its rank and suit, and determine the game outcome in real-time[cite: 15, 16].

The core goal was to prove that classic image processing algorithms (like Template Matching) can solve real-world object recognition problems efficiently without needing heavy deep learning models[cite: 37].

[Watch the System Demonstration [cite: 155]]

🛠️ Tech Stack

• Language: Python
• Vision Library: OpenCV (Open Source Computer Vision Library) [cite: 60]
• Game Engine: Pygame [cite: 109]
• Hardware: Standard Webcam, Laptop

⚙️ How It Works: The Vision Pipeline

The system is split into two main modules: the Vision Engine (perception) and the Game Logic[cite: 102]. Here is the step-by-step process of how the computer "reads" a card:

1. Pre-processing & Edge Detection

Before analyzing the card, the image must be simplified to reduce computational load[cite: 63].

• Grayscale Conversion: Converting the image from BGR (Blue-Green-Red) to a single intensity channel.
• Gaussian Blur: Smoothing the image to reduce high-frequency noise[cite: 64].
• Canny Edge Detection: Extracting the structural lines of the card by detecting drastic changes in pixel intensity (gradients)[cite: 66, 67].

2. Perspective Transform (The "Homography" Step)

A major challenge in computer vision is that cards are rarely held perfectly straight. I implemented a Perspective Transform to fix this[cite: 71].

• The system detects the four corners of the card contour.
• It calculates a transformation matrix to "warp" the image.
• The result is a flattened, top-down view of the card, regardless of how the user holds it[cite: 72, 75].

3. Template Matching (Classification)

Instead of using Neural Networks, I used a lightweight Template Matching method with the Sum of Absolute Differences (SAD) algorithm[cite: 17, 78].

• The system compares the flattened card image against a database of 52 reference card images.
• The template with the lowest difference score is identified as the match[cite: 79, 80].

🎲 Game Logic Implementation

The game follows the standard "Higher or Lower" rules implemented in game_logic.py[cite: 106].

• Mapping: Cards are mapped to numerical values (e.g., Jack = 11, King = 13)[cite: 121].
• State Management: The system handles states to prevent crashes, ensuring a card is detected before a guess is registered[cite: 122].

# Snippet: Safety validation before processing a guess [cite: 124-129]
if self.current_card is None:
    self.message = "Wait for Reference Card! Press 'R'."
    return False
if self.next_card is None:
    self.message = "Wait for GUESS Card detection!"
    return False