🛡️ SafeTurn AI

Intelligent Free-Left Turn Traffic Signal Controller

"Predictive, not reactive — we act before danger arrives."

Built for HackSETU 2025 | Theme 3: Intelligent Free-Left Turn Management for Safer Roads

---

💡 What It Does

SafeTurn AI uses a camera feed with YOLOv8 real-time object detection to count pedestrians at a junction and dynamically control traffic signals for free-left turns. The system implements a realistic 3-stage signal (🟢 GREEN → 🟡 ORANGE → 🔴 RED) with stability logic to prevent flickering.

    ┌──────────────────────────────────────────────┐
    │           SafeTurn AI Pipeline                │
    │                                               │
    │  Camera Feed (video or webcam)                │
    │     │                                         │
    │     ▼                                         │
    │  YOLOv8n Detection (person, car, bike, bus)   │
    │     │                                         │
    │     ▼                                         │
    │  Pedestrian Count                             │
    │     │                                         │
    │     ▼                                         │
    │  3-Stage Signal Controller                    │
    │     │                                         │
    │     ▼                                         │
    │  Decision: 🟢 GREEN │ 🟡 ORANGE │ 🔴 RED     │
    └──────────────────────────────────────────────┘

---

🚦 Signal Logic

Pedestrians	Action
0	🟢 Stay GREEN — free turn
1	Wait 5 seconds, then 🟡 ORANGE → 🔴 RED
2+	Immediately 🟡 ORANGE → 🔴 RED

Transition Rules

• GREEN → ORANGE → RED (never skips a stage)
• 🟡 ORANGE duration = 2 seconds (warning)
• 🔴 RED minimum = 3 seconds
• RED → GREEN only when 0 pedestrians detected

---

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run the System

# Single camera — one video with pedestrians + vehicles
python app.py --mode single --video test4vedio.mp4

# Single camera — webcam
python app.py --mode single --camera 0

# Dual camera — webcam (pedestrians) + video (traffic)
python app.py --mode dual --video test4vedio.mp4

Controls

Key	Action
Q	Quit
S	Save screenshot
R	Rewind video (single mode)
A	Simulate accident

---

🏗️ Architecture (Modular)

app.py                    ← Main entry point (--mode single / dual)
├── detection.py          ← YOLOv8 wrapper with frame resize optimization
├── signal_controller.py  ← 3-stage state machine (GREEN → ORANGE → RED)
│
├── Single Mode           ← One camera: detects persons + vehicles
└── Dual Mode             ← Webcam (pedestrians) + Video (traffic view)

Other Files

safeturn_final.py         ← Legacy complete system (predictive engine)
safeturn_main.py          ← Legacy single-camera with zone detection
step2_dual_camera.py      ← Standalone single-camera version
dashboard.py              ← Streamlit monitoring dashboard

---

⚡ Performance Optimizations

Optimization	Details
Frame skipping	YOLO runs every 2nd frame, cached results reused
Frame resize	Input resized to 640×360 before inference
YOLO imgsz	Internal YOLO size set to 480px (vs default 640)
YOLOv8n	Nano model — fastest, optimized for CPU
cv2.waitKey(1)	Zero display delay for smooth playback

---

🎯 Detection & Visualization

Detected classes with color-coded bounding boxes:

Class	Color	COCO ID
person	🟢 Green	0
car	🔴 Red	2
bicycle / motorbike	🟠 Orange	1, 3
bus	🩷 Pink	5
truck	🫧 Teal	7

Each detection shows: class name + confidence (e.g., person 85%)

---

🚨 Accident Simulation

Press A during the demo to trigger:

Feature	What Happens
🔴 Signal Override	Immediate RED for 5 seconds
🚑 Ambulance Alert	"Ambulance Alert Triggered" displayed
⚠️ Visual	Flashing red border + center banner

---

🛠️ Tech Stack

• Python — core language
• OpenCV — video processing + UI overlays
• YOLOv8 (ultralytics) — real-time object detection
• NumPy — array operations

---

👥 Team

Built at HackSETU 2025