Traffic Flow Analyzer

The Traffic Flow Analyzer is a deep learning-powered tool designed for real-time monitoring and analysis of vehicle traffic. Leveraging the YOLOv11x object detection model, this project detects vehicles and tracks their movement using the BoT-SORT tracker. It estimates pixel-based speed for each detected vehicle using Euclidean distance between successive frames. The tool provides key insights such as the total number of vehicles, the count of each vehicle type/class, and identifies congestion hotspots based on vehicle density and average speed. This makes it an ideal solution for urban planners, transport authorities, and smart city enthusiasts focused on improving traffic management and road safety.

---

Features

• Real-time vehicle detection and tracking using YOLOv11x and BoT-SORT.
• Pixel-based speed estimation for each detected vehicle using Euclidean distance.
• Insights on total vehicle count, vehicle counts per class, and congestion hotspots.
• Congestion status determined based on vehicle density and average speed.

---

Repository Structure

• Congestion_detection/: Contains the main Python script for congestion detection and YOLO model weights.
• Model yolo11x/: Stores the training script.
• train.py: Script to train the YOLOv11_x object detection model.
• dataset.txt: Provides links to traffic datasets and annotations.

---

Setup Instructions

1. Clone the Repository

git clone https://github.com/The-Red-Wood-Lab/Smart-Traffic-Flow-Analyzer.git
cd Smart-Traffic-Flow-Analyzer

2. Install Dependencies

Ensure Python 3.7+ is installed, then run:

pip install -r requirements.txt

---

Run the Detection

GPU Requirements

• Running the main.py script efficiently requires a GPU.
• The script was tested on Google Colab with a T4 GPU.

1. Navigate to the Congestion_detection Directory

cd Congestion_detection

2. Run the Main Script

python main.py --weights path/to/weights.pt --input path/to/your_video.mp4 --output path/to/output_video.mp4

Arguments:

• --weights: Path to the YOLO model weights (e.g., model/yolo_11x_traffic.pt).
• --input: Path to the input video for analysis.
• --output: Path to save the output video (default: output_video.mp4).

Example Usage

python main.py --weights model/yolo_11x_traffic.pt --input traffic_video.mp4 --output detected_output.mp4

---

Output Example

The output video output.mp4 is saved in the Congestion_detection folder. This video contains annotated frames highlighting detected congestion areas.

Training the Model

train.py Script

The train.py script in the model/ directory was used to train the YOLOv11_x object detection model.

Training Environment

• The model was originally trained on Kaggle using 2x T4 GPUs.
• The dataset links are provided in the dataset.txt file for reference and reproducibility.

Key Training Details

• The model is fine-tuned for vehicle detection in traffic scenarios.
• Pre-trained weights and additional resources can be added to the repository for convenience.

---

Dataset

The dataset file contains essential links to traffic footage and annotations used for training and testing the deep learning models. Ensure you download and use these datasets for further analysis or contributions.

---

Additional Tips

• Google Colab: If you plan to run the project on Colab, ensure you enable GPU in the runtime settings for optimal performance.
• Custom Input: Replace the --input path with your video file for personalized analysis.
• Kaggle: Alternatively, you can run the project on Kaggle. The notebook is optimized for use on Kaggle Kernels, where you can take advantage of Kaggle's free GPU resources.

---