Vehicle Speed Estimation

Overview

This project uses the optical flow algorithm, specifically the Lucas-Kanade tracker, to estimate vehicle speeds from mono camera (CCTV) footage.

Prerequisites

• Python 3.x
• Required libraries: opencv-python, numpy

Setup

1. Clone the Repository

   git clone https://github.com/swhan0329/vehicle_speed_estimation.git
   cd vehicle_speed_estimation

2. Install Dependencies

   pip install -r requirements.txt

Usage

1. With an Input Video

   python main.py [input video name]

   Replace [input video name] with the path to your video file.

2. Without an Input Video The script will automatically use the webcam on your computer.

   python main.py

3. Additional Options

   python main.py [input video name] --output output.mp4 --show

File Descriptions

• main.py: The main script to run the vehicle speed estimation.
• video.py: Contains functions to handle video input and processing.
• common.py: Includes common functions and utilities used across the project.
• tst_scene_render.py: Test and render scenes for visualization.

Example

To run the speed estimation on a sample video:

python main.py sample_video.mp4

To use the webcam for live speed estimation:

python main.py

Additional Notes

• Ensure that your video has a clear view of the road and vehicles for accurate speed estimation.
• Adjust parameters in common.py if needed to fit specific requirements or to improve performance.
• The road-region settings are hardcoded for a specific camera view in main.py:
  • view_polygon
  • cal_polygon
  • polygon1 ~ polygon5 (lane areas)
• These coordinates must be re-collected and re-tuned manually for each new road/camera angle/resolution.
  If not updated, lane assignment and speed estimates can be incorrect.
• Pixel-to-meter scale values are also hardcoded (px2m1 ~ px2m5) and should be recalibrated per camera setup.

Running Tests

Execute the unit tests with Python's built-in test runner:

python -m unittest discover -s tests