Collision Prediction System

A Real-time Collision Prediction Alert System Using Monocular Cameras

Background

• The commercialization of delivery robots utilizing autonomous driving technology has significantly increased the demand for depth estimation technology.
• Traditional distance estimation is possible with Lidar sensors, but they are expensive.
• Videos captured on the streets often include people’s faces, making it difficult to use such data.

Expected Benefits

• Prevent safety accidents through distance measurement technology using monocular cameras.
• Overcome the limitations of monocular cameras in adverse weather conditions using deep learning.
• Reduce costs by using monocular cameras instead of expensive Lidar sensors.

My Role

• Implementation of Python GUI app.
• Implementation of real-time model serving server.

Models Used

• YoloWorld
• Depth Anything

Project Structure

Technologies Used

• PYQT5
• OpenCV
• WebSockets
• FastAPI
• Lock
• ONNX
• Pytorch

Project Link

GitHub Repository