Traffic Analysis Project

This project provides tools and scripts for analyzing traffic from videos. It uses computer vision techniques and deep learning models to detect vehicles and measure their speeds.

Features

• Vehicle Detection: Identify and classify vehicles in various lanes.
• Bird's-Eye View Transformation: Transform the video frame to get a top-down view, making it easier to analyze traffic patterns.
• Lane Detection: Automatically detect lanes to segregate vehicles.
• Speed Measurement: Calculate the speed of each detected vehicle.
• Data Aggregation: Summarize and group detected data for easier analysis.
• Integration with Transformers Library: Utilizes pretrained models from Hugging Face's model hub for object detection.

Prerequisites

Make sure you have the following libraries installed:

pip install opencv-python matplotlib transformers timm torch torchvision

Usage

1. Point Selection: The script starts by allowing users to select four points on the first frame of the video. These points are used to transform the video frame to get a bird's-eye view.

2. Frame Processing: Each frame is then processed to detect vehicles, measure their speeds, and identify the lanes they are in.

3. Data Aggregation: After processing all frames, the script aggregates the data to give a summary of the number of vehicles detected for each type and in each lane.

4. Output: The final data is presented in a pandas DataFrame format, showing aggregated data for each timestamp, label (vehicle type), and lane.

Code Structure

1. Video Loading and Initialization: Loads the video and initializes necessary parameters.
2. Mouse Callback for Point Selection: Allows users to select points on the video frame.
3. Bird's-Eye View Transformation: Transforms the video frame to get a top-down view using the selected points.
4. File Operations: Includes merging frames from different sources and creating an aggregated dataset.
5. Annotation Parsing: Parses annotation files to identify objects in frames.
6. Model Initialization and Training: Loads pretrained DETR model, fine-tunes it on the dataset, and pushes it to Hugging Face's model hub.
7. Object Detection in Video: Processes each video frame to detect vehicles, measure their speeds, and identify the lanes they are in.
8. Data Aggregation and Output: Aggregates the detected data and outputs it in a pandas DataFrame format.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT