Performance Evaluation of an Object Detection Model Using Drone Imagery in Urban Areas for Semi-Automatic Artificial...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Urban Planning

Background:

• Deep learning models are crucial for image processing in traffic monitoring.
• Developing these AI models requires extensive, resource-intensive training data.
• AI-based urban management faces challenges due to data creation bottlenecks.

Purpose of the Study:

• To propose an efficient method for constructing AI training datasets for semi-moving object detection.
• To analyze tasks for refining AI model outputs in urban traffic monitoring.
• To reduce resource requirements in AI model development.

Main Methods:

• Utilized an existing AI object detection model for training data construction.
• Analyzed refinement tasks for AI-detected objects.
• Evaluated performance using F-Beta scores (F0.5, F1, F2).

Main Results:

• The F2 score demonstrated superior performance in improving dataset completeness.
• Achieved 26.5% less effort than F0.5 and 7.1% less effort than F1.
• Proposed an efficient evaluation method for semi-automatic AI training data construction.

Conclusions:

• The proposed method enables efficient creation of AI training data for urban traffic monitoring.
• Optimizing evaluation metrics like F2 score significantly reduces resource demands.
• Facilitates future AI model development for AI-based urban environment management.