This project is a complete end-to-end traffic intelligence system built using Computer Vision + Transformers + LLMs. It processes a traffic video and automatically produces:

✔ Vehicle detections ✔ Multi-object tracking with stable IDs ✔ Vehicle counting ✔ Speed estimation (km/h) ✔ Traffic light (red/green) detection ✔ Red-light violation logs ✔ CSV/JSON event data ✔ Human-readable traffic report generated by an LLM

Demo Video

Event Log (CSV Output)

AI-Generated Traffic Report

Summary of Observations:

• Total Crossings: 11
• Red-Light Violations: 5
• Average Speed: 71.2 km/h
• Maximum Speed: 104.7 km/h

Traffic Density: The traffic density appears to be moderate, with a total of 11 crossings recorded during the observation period. This level of density suggests a manageable flow of vehicles, but the presence of multiple violations indicates potential issues with driver compliance and awareness.

Speed Behavior Assessment: The average speed of 71.2 km/h is above typical urban speed limits, which often range from 50 to 60 km/h. The maximum recorded speed of 104.7 km/h is particularly concerning, as it significantly exceeds safe driving limits, especially in areas with traffic signals. This behavior poses a risk not only to the drivers but also to pedestrians and other road users.

Violation Commentary: A total of 5 red-light violations were recorded, representing a significant 45% of total crossings. These violations occurred under a red light condition, indicating a serious disregard for traffic signals. The speeds at which these violations occurred ranged from 52.4 km/h to 100.5 km/h, further exacerbating the risk of accidents. The high rate of violations suggests a need for increased enforcement and public awareness regarding traffic laws.

Safety Recommendations:

1. Increased Enforcement: Implement targeted enforcement measures at the intersection to deter red-light violations. This could include the use of automated traffic cameras and increased police presence during peak hours.

2. Public Awareness Campaigns: Launch educational campaigns to inform drivers about the dangers of speeding and running red lights. Emphasizing the legal consequences and potential accidents could help change driver behavior.

3. Traffic Signal Improvements: Consider upgrading traffic signal visibility and timing, as well as installing additional signage to alert drivers of upcoming signals. Enhanced visibility may help reduce violations and improve overall traffic safety.

📌 Key Features

1. RT-DETR Transformer-Based Detection

• High-accuracy transformer detector
• Detects: car, truck, bus, motorbike, bicycle

2. Custom Object Tracking

A lightweight, frame-to-frame centroid tracker for stable IDs.

3. Vehicle Counting

A virtual line detects crossing events.

4. Speed Estimation

• Pixel distance per frame
• Converts to km/h
• Logged into CSV

5. Red-Light Violation Detection

• Detects traffic light color (RED/GREEN)
• Logs violations automatically

6. Structured Event Logging

Exports:

results/
├── traffic_events_core.csv
└── traffic_events_core.json

Each log entry contains:

• Frame
• Timestamp
• Vehicle ID
• Vehicle class
• Speed
• Event type
• Light state

🧠 7. LLM-Powered Traffic Summary

Using gpt-4o-mini, the system generates a professional traffic analysis report.

🏗 Project Structure

traffic-analysis-ai/
├── src/
│   ├── traffic_core.py      # Vision pipeline (video → logs)
│   └── llm_report.py        # LLM report generator
├── data/                    # Input videos (ignored in Git)
├── results/                 # Output video + event logs (ignored)
├── requirements.txt
├── .env                     # OpenAI key (not committed)
├── .gitignore
└── README.md

🛠 Tech Stack

🚀 Getting Started

1️⃣ Clone Repository

git clone https://github.com/<your-username>/traffic-analysis-ai.git
cd traffic-analysis-ai

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Add Your Traffic Video

Place your input video here:

data/traffic.mp4

(or update the path in src/traffic_core.py).

4️⃣ Add Your OpenAI API Key

Create a .env file in the project root:

echo "OPENAI_API_KEY=your_openai_key_here" > .env

.env is ignored by Git for security.

🎬 Run Traffic Analysis

python src/traffic_core.py

This will generate:

• results/output_core.mp4
• results/traffic_events_core.csv
• results/traffic_events_core.json

🧠 Generate LLM Traffic Report

python src/llm_report.py

Sample console output:

Traffic during the analyzed period was moderately dense...
Average speeds remained within a safe range...
One red-light violation was detected...
Recommended improvements include...

📈 Sample Log Output (CSV)

frame,time_sec,track_id,class_name,speed_kmh,event_type,light_state
109,3.63,2,car,42.1,CROSSING,GREEN
215,7.16,7,truck,38.6,RED_LIGHT_VIOLATION,RED
332,11.06,5,bus,29.8,CROSSING,GREEN

🧩 System Architecture

┌────────────────┐
│ traffic.mp4 │
└──────┬─────────┘
│
┌───────▼────────┐
│ RT-DETR │
└───────┬────────┘
│ detections
┌───────▼──────────┐
│ Simple Tracker │
└───────┬──────────┘
│
┌───────▼──────────────┐
│ Speed Estimator │
└───────┬──────────────┘
│
┌───────▼──────────────────┐
│ Event Logger (CSV/JSON) │
└───────┬──────────────────┘
│
┌───────▼───────────────────────┐
│ gpt-4o-mini LLM Report │
└───────────────────────────────┘

🤝 Contributing

Pull requests are welcome. If you find a bug or have an idea, feel free to open an issue.