🚦 TraffiQ — AI-Powered Traffic Analysis System

A professional-grade real-time traffic analysis system built with YOLOv8, ByteTrack, and Flask. Detects and tracks vehicles across user-defined lanes, computes per-lane statistics, flags incidents, measures speeds, and displays everything in a live web dashboard with a purple-black gradient UI.

Featuring a premium, two-way interactive Control Panel allowing live parameter configuration without restarting the system.

---

✨ Features

Feature	Description
⚙️ Interactive Control Panel	Collapsible sidebar for live web-based parameter configuration
🎯 Multi-Object Tracking	ByteTrack algorithm with unique vehicle IDs
🛣️ Auto Lane Mapping	Lanes mapped from vehicle trajectories (vanishing-point wedges), Y/N confirm, manual 4-click fallback
📐 Auto Speed Calibration	Camera height/horizon from car sizes, depth scale from lane dashes — no manual pixel-to-metre
⚡ Lane-Level Analytics	Per-lane counts, Level of Service (LOS) A–F grading, flow rate, queue length
📷 Speed Camera	Auto-calibrated ground-plane speed estimation + dynamic speeding cameras + CSV logging
⚠️ Incident Detection	Dynamic threshold incident tracking for stopped vehicles (>2s to 30s)
🚑 Ambulance Detection	Open-vocabulary YOLO-World ("ambulance" prompt) confirms ambulances on tracked vehicles and pre-empts the signal
🚦 Actuated Signal Control	Min green, gap-out on line-crossing detectors, demand-scaled max green, yellow clearance, starvation guard
⏱️ Accurate Wait Times	Red-lane waits predicted with the same priority rule the controller applies
➖ Line-Crossing Counts	Per-lane count line (travel axis learned from motion) with N/S/E/W direction totals
🧭 Vehicle Heading	Per-vehicle compass heading; EMA-smoothed speeds
📊 Live Chart History	2-minute rolling bar/line chart (Chart.js)
🔥 Heatmap Overlay	Accumulative density heatmap via OpenCV
📈 Trend Prediction	Rolling linear regression per lane
💾 CSV Data Logging	Every detection and violation exported for offline analysis
🔐 Auth-Protected Dashboard	Login-gated live dashboard with session management

---

⚙️ Interactive Control Panel & Sidebar

TraffiQ transitions from a passive monitor to an active control center via its dynamic two-way Control Panel:

• Collapsible Sidebar: Open and close the control console via the hovering gear (⚙️) button or Escape key. Styled with blur backdrops and neon-purple glow themes.
• Visualization Mode Toggles: Switch between active processing modes (Lanes, Heatmap, Speed, Timer) dynamically. The underlying OpenCV pipeline renders the corresponding overlay instantly.
• Dynamic Speed Trigger Limit: Adjust speeding camera threshold (20 - 120 km/h) on the fly via slider control. Bounding boxes highlight speeding vehicles instantly.
• Incident Timeout Configuration: Slide the stopped vehicle detection threshold (2s - 30s) dynamically.
• YOLO Confidence Slider: Slide detection threshold (0.10 - 0.95) to filter out background noise or increase object detection density instantly.
• Intelligent Cooldown Lockouts: Features a robust 5-second interaction cooldown protection. It displays an amber-glass warning banner reading Please wait 5 seconds before you change when inputs are spammed, and automatically reverts sliders/modes to verified backend states on failure.

---

🚀 Getting Started

1. Clone the repo

git clone https://github.com/Varn1t/Traffic-Intelligence.git
cd "Traffic Intelligence"

2. Install dependencies

pip install -r requirements.txt

3. Set your credentials

Copy the example env file and fill in your own values:

cp .env.example .env

Edit .env:

DASH_USER=your_username
DASH_PASS=your_password

4. Add your video file

Place your traffic video in the project folder and update VIDEO_PATH in config.py:

VIDEO_PATH = "Traffic_Video.mp4"

5. Run the System

python main.py

1. Auto lane mapping (~15 s): TraffiQ watches the first AUTO_LANE_FRAMES frames, tracks vehicles, and maps each lane from where traffic actually drives (lanes are wedges meeting at the road's vanishing point, so they follow perspective).
2. "Are the lanes correctly mapped?" Press Y to continue, or N to draw them yourself (click the 4 corners of each lane, U to undo, ENTER when done).
3. Speed calibration is automatic: camera height and horizon are learned from the sizes of passing cars, and the depth scale from dashed lane markings (DASH_CYCLE_M, 12 m for US/Canada highways). It keeps refining while running; the HUD shows CAL: cam …m f …px (dashes|cars|guess).
4. The YOLOv8 model runs in a background thread, starts local camera playback, and spins up the live dashboard.
5. Open http://localhost:5050 in your browser, log in, and control the system dynamically.

---

🛠️ Tech Stack

• Python — core logic
• YOLOv8 (Ultralytics) — vehicle detection
• YOLO-World (Ultralytics) — zero-shot ambulance detection
• ByteTrack (Supervision) — multi-object tracking
• OpenCV — video processing & heatmap overlays
• Flask — web server & REST API
• Chart.js — live dashboard charts
• scikit-learn — Random Forest traffic prediction
• Vanilla CSS — premium glassmorphism styling and animations

---

📁 Project Structure

traffiq/
├── main.py             # Main entry point & video processing loop
├── config.py           # Configuration variables and thresholds
├── state.py            # Thread-safe shared state & stats tracking
├── loggers.py          # Logging classes (CSV, SpeedCamera)
├── analytics.py        # Core logic: ML predictor, incident detection, flow rates
├── templates.py        # UI: Dashboard and landing page HTML
├── web_app.py          # Flask application logic
├── requirements.txt    # Python dependencies
├── .env.example        # Credential template (copy → .env)
├── .gitignore
└── README.md

Note: Your .env, video files (*.mp4), model weights (*.pt), and CSV logs are excluded from the repo via .gitignore.

---

📄 License

All rights reserved by Varn1t

---

Built by Varnit