ENSI Computer Vision System - Eurobot 2025

Overview

This repository contains our complete Computer Vision system for the ENSI Eurobot 2025 competition. The system successfully implements vision-based solutions for all three competition tasks:

• Task 1: Color Detection - Identify red and blue objects with size filtering
• Task 2: Number Recognition - Detect digits 3, 5, 6, 9 using fine-tuned YOLO
• Task 3: Object Detection - Identify green cubes and triangles with verification

The system evolved through two major versions, with the final implementation using a unified YOLOv8-based approach that achieved competition-winning performance.

---

Repository Structure

Computer_vision/
├── README.md                           <- Project documentation
├── requirements.txt                    <- Python dependencies
├── resources.md                        <- Computer vision resources and tutorials
├── first_version(yolo+model_keras+opencv)/
│   ├── README.md                       <- Detailed first version analysis
│   ├── task1_color_detection/          <- HSV-based color detection
│   ├── task2_number_detection/         <- Keras CNN + OpenCV approach
│   └── task3_object_detection/         <- YOLO + complex post-processing
└── second_version(finetuned_yolo_for_everything)/
    ├── README.md                       <- Production system documentation
    ├── computer_vision_service.py      <- ROS service integration
    ├── task1_color_detection/          <- Improved color detection
    ├── second_task_number_detection/   <- Fine-tuned YOLO for digits
    └── task3_object_detection/         <- Optimized object detection

---

🏆 System Versions

First Version (Hybrid Approach)

• Location: first_version(yolo+model_keras+opencv)/
• Approach: Mixed technologies (OpenCV + Keras + YOLO)
• Issues: Poor number detection accuracy (~60%), complex pipelines, multiple windows
• Status: ❌ Learning prototype with identified limitations

Second Version (Unified YOLO)

• Location: second_version(finetuned_yolo_for_everything)/
• Approach: Fine-tuned YOLOv8 for all tasks
• Performance: Excellent accuracy (~95% for numbers), robust outdoor operation
• Status: ✅ Competition-ready production system

---

🚀 Quick Start

Prerequisites

pip install ultralytics opencv-python numpy rospy

Running the Production System (Second Version)

Individual Tasks

# Color Detection
cd second_version(finetuned_yolo_for_everything)/task1_color_detection
python color_detection_ver1.py

# Number Detection  
cd second_version(finetuned_yolo_for_everything)/second_task_number_detection
python yolo_only.py

# Object Detection
cd second_version(finetuned_yolo_for_everything)/task3_object_detection/last_version  
python yolo_minimal.py

ROS Service Integration

# Start computer vision services
cd second_version(finetuned_yolo_for_everything)
python computer_vision_service.py

# Call services from robot
rosservice call /service1  # Color detection
rosservice call /service2  # Number detection  
rosservice call /service3  # Object detection

---

🛠️ Technology Stack

Core Technologies

• YOLOv8 (Ultralytics) – Primary detection engine for all tasks
• OpenCV – Image preprocessing, color filtering, camera interface
• ONNX Runtime – Optimized model inference for edge deployment
• ROS – Robot communication and service integration

Processing Pipeline

• CLAHE – Adaptive contrast enhancement for sunlight robustness
• Gaussian Blur – Noise reduction preprocessing
• HSV Color Filtering – Object color verification
• Confidence Thresholding – Best detection selection

---

📊 Performance Achievements

Task	First Version	Second Version	Improvement
Color Detection	HSV filtering	HSV + size constraints	✅ Better reliability
Number Recognition	~60% (Keras CNN)	~95% (Fine-tuned YOLO)	✅ +35% accuracy
Object Detection	Complex post-processing	Clean single detection	✅ 2x faster
Sunlight Robustness	Poor	Excellent	✅ Outdoor ready
Code Maintainability	Mixed technologies	Unified YOLO	✅ Much simpler

---

🏁 Competition Results

✅ Successfully implemented all three required computer vision tasks
✅ Robust outdoor performance with CLAHE preprocessing
✅ Real-time detection suitable for competition constraints
✅ ROS integration for seamless robot communication
✅ Production deployment ready for competition environment

---

📚 Documentation

• First Version README - Detailed analysis of initial approach and lessons learned
• Second Version README - Production system documentation and technical details
• Resources - Computer vision learning materials and references

---

🤝 Development Team

INSAT Eurobot 2025 Team - Computer Vision Module

For questions or contributions, please contact the development team or create an issue in this repository.

---

📄 License

This repository is for Eurobot 2025 INSAT Team use only. External usage requires explicit permission.