Vision-Driven Card Prejudgment System for Live Shuffling Analysis

Here is the professional translation of your school project assignment into English:

Vision-Based Card Prejudgment System
Demonstrated in Python (Other platforms acceptable for implementation)

I. Project Objectives
Develop a Python program to achieve the following core functionalities in YouTube Live scenarios using computer vision:

Live video monitoring

Hand and card detection

Card face recognition (UP/Bomb)

Card tracking and sequence recording

The system must predict card types and record their final order (1-10) before they are covered by a bowl during the teacher's live shuffling process (10 rounds).

Note: Detection and prediction must be completed before cards are revealed.

II. Rules
Card Setup:

10 cards (5 UP, 5 Bomb) with identical backs and distinct fronts.

Procedure:

The teacher shuffles cards live for 10 rounds, then stacks them as positions 1-10 under a bowl.

Full marks for 10 correct predictions, passing for 6, and zero for ≤5.

Each round accommodates 5 students.

Key Requirement:
The system must predict card types through frame-by-frame analysis before the bowl covers them.

III. Technical Requirements
Core Modules

Hand Detection

Locate teacher’s hands to narrow card detection areas.

Card Detection & Tracking

Real-time position detection and trajectory tracking.

Front/Back Classification

Detect card flips and classify fronts as UP/Bomb.

Sequence Recording

Output card positions (1-10) and types after shuffling.

Performance Standards

Real-time: ≥15 FPS (matching live stream rates).

Robustness: Resistant to hand occlusion, lighting changes, and fast card movements.

IV. Deliverables
Code Submission

Complete Python project (models, dependencies).

Well-commented code with algorithm rationale.

Test Report

Demo video: Show shuffling detection, classification, and tracking.

Metrics: Accuracy (≥90%), FPS.

Innovation (Bonus):
E.g., improved tracking algorithms or optimized classification models.

V. Recommended Tools & Stack
Frameworks: OpenCV (video), PyTorch/TensorFlow (ML).

Models: YOLO (detection), DeepSORT (tracking), MobileNet (classification).

Optimization: Multithreading/GPU acceleration.

VI. Grading Rubric
Module Points Criteria
Hand/Card Detection 30 ≥85% accuracy, reproducible code
Front Classification 30 ≥90% UP/Bomb accuracy
Tracking & Sequencing 25 Correct 1-10 type mapping
Innovation & Report Quality 15 Clear logic, optimization insights
Deadline: May 23, 23:59
Submission: To be determined

This translation preserves technical terminology while adhering to academic/professional standards. Let me know if you need further refinements!