UAV Based Indoor Localization and Objection Detection.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Unmanned Aerial Vehicles (UAVs) require precise indoor positioning and real-time 3D target detection for complex tasks.
• Existing methods often struggle with accuracy and speed in dynamic indoor environments.

Purpose of the Study:

• To develop and verify a novel method for fast indoor positioning and 3D target detection for UAVs.
• To enable real-time task implementation through accurate localization and semantic understanding of the environment.

Main Methods:

• Combined direct and feature-based visual odometry for camera pose estimation using photometric error.
• ORB features and Bundle Adjustment (local and global optimization) for map consistency.
• Depth filtering for map point convergence and Convolutional Neural Network (CNN) with YOLOv3 for target detection.
• Association of spatial map points with target detection boxes and statistical outlier filtering for noise reduction.

Main Results:

• The proposed method achieves fast and accurate UAV position estimation in indoor environments.
• Effective 3D target detection and semantic information extraction using YOLOv3.
• Successful integration of localization and object detection for real-time UAV applications.

Conclusions:

• The integrated approach significantly enhances UAV capabilities for indoor navigation and object recognition.
• The method demonstrates efficacy through experiments on public datasets and a custom UAV platform.
• This research contributes to advancing autonomous indoor operations for UAVs.