Unmanned Aerial Vehicles (UAVs) play an important role in very precise navigation for its usage in applications like disaster management, agricultural surveillance, and as delivery drones. Precise depth estimation is crucial for UAV to safely avoid obstacles and thus obtain proper paths for flying. Monocular Depth Estimation (MDE), is another method which is becoming popular due to its easy implementation as compared to stereo vision and LiDAR systems in which single camera depth information is inferred. However, found that there are still some problems in the current a few deep learning methods of MDE, including high computational cost, low prediction accuracy in dynamic conditions, and unsuitability for real-time or online applications. In order to solve these problems, the current research work introduces a Segmental Fully Convolutional Neural Network (SFCNN) tended by the Prairie Dog Optimization (PDO) for UAV flight. The block structure improves adaptability as well as computational complexity; PDO seeks to minimize network's parameters to increase the result precision of the depth forecast and lessen the computation time. tarting from this background, the proposed approach is intended to support lightweight real-time and reliable means for MDE to enhance UAV navigation in complex environments.
Monocular Depth Estimation for UAV Navigation Using Segmental Fully Convolutional Prairie Dog Neural Networks
2025-02-18
482908 byte
Conference paper
Electronic Resource
English
An Improved Convolutional Neural Network for Monocular Depth Estimation
British Library Conference Proceedings | 2020
|An Improved Convolutional Neural Network for Monocular Depth Estimation
Springer Verlag | 2020
|