Vehicle Detection in Overhead Satellite Images Using a One-Stage Object Detection Model

Area of Science:

• Computer Science
• Artificial Intelligence
• Remote Sensing

Background:

• Urban congestion necessitates advanced traffic monitoring and prediction methods.
• Satellite and drone imagery offer valuable data for understanding vehicle behavior.
• Machine learning models are crucial for extracting insights from overhead imagery.

Purpose of the Study:

• To develop and present a one-stage object detection model for identifying vehicles in satellite imagery.
• To evaluate the model's performance in terms of accuracy and detection speed.

Main Methods:

• Utilized the RetinaNet architecture for object detection.
• Trained and tested the model on the Cars Overhead With Context dataset.
• Focused on extracting vehicle count, position, and direction information.

Main Results:

• Achieved high accuracy in vehicle detection.
• Demonstrated a very low detection time, indicating real-time processing capability.
• The model effectively extracts crucial data from overhead images.

Conclusions:

• The proposed RetinaNet-based model is effective for real-time vehicle detection in satellite images.
• This technology can significantly contribute to improving urban traffic management and congestion avoidance.
• The model's efficiency makes it suitable for deployment with live satellite or drone data feeds.