A Path Planning Method with Perception Optimization Based on Sky Scanning for UAVs.

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Disaster Management Technology

Background:

• Unmanned aerial vehicles (UAVs) are crucial for disaster management, providing valuable aerial perspectives.
• Current UAV operations are hindered by image instability and insufficient data for deep learning models, impacting decision-making accuracy.
• Multiple image captures from various viewpoints are often necessary for reliable object recognition.

Purpose of the Study:

• To formulate an optimal UAV flight path task that addresses real-world perception requirements in disaster scenarios.
• To enhance the effectiveness and efficiency of UAVs in scene understanding and object recognition during disaster response.
• To minimize time costs while maximizing object recognition capabilities for UAVs.

Main Methods:

• Development of a convolutional neural networks (CNNs) model for object detection and localization (e.g., buildings).
• Implementation of an optimization method to determine the most efficient flying path for comprehensive scene recognition.
• Integration of perception needs into the UAV path planning process.

Main Results:

• Simulations demonstrated the proposed method's effectiveness in addressing real-world scene understanding challenges for UAVs.
• The optimized path planning approach proved efficient, reducing time costs for object recognition.
• The CNN model successfully detected and localized key objects, improving data quality for decision-making.

Conclusions:

• The proposed CNN and path optimization method significantly enhances UAV capabilities in disaster management.
• This approach effectively balances perception needs with efficient path planning for real-world applications.
• The study provides a robust solution for improving UAV-based scene understanding and decision-making in critical situations.