Related Experiment Video
Updated: Jul 13, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
568
Monocular catadioptric panoramic depth estimation via improved end-to-end neural network model
Fei Yan1,2, Lan Liu1, Xupeng Ding1
1School of Electronic Information Engineering, Changchun University of Science and Technology, Changchun, China.
Frontiers in Neurorobotics
|October 11, 2023
Summary
This study introduces an improved neural network for panoramic depth estimation using catadioptric cameras. The enhanced algorithm achieves highly accurate object depth estimates, improving computer vision applications.
Area of Science:
- Computer Vision
- Machine Learning
- Robotics
Background:
- Accurate depth estimation is crucial for autonomous systems.
- Monocular catadioptric cameras offer wide field-of-view imaging.
- Existing depth estimation methods face challenges with panoramic data.
Purpose of the Study:
- To develop a robust monocular catadioptric panoramic depth estimation algorithm.
- To improve the accuracy and reliability of depth estimation from panoramic images.
- To leverage advanced neural network architectures for enhanced image understanding.
Main Methods:
- An improved end-to-end neural network model was developed.
- An enhanced concentric circle approximation unfolding algorithm was used for image processing.
- Non-local attention mechanisms and depth smoothness loss were integrated.
Main Results:
- The algorithm effectively unfolds and processes panoramic images.
- Non-local attention improved image understanding capabilities.
- Depth smoothness loss enhanced the precision and reliability of depth estimates.
Conclusions:
- The proposed algorithm provides highly accurate object depth estimates.
- This method advances depth estimation for monocular catadioptric systems.
- The refined approach has significant potential for real-world applications.
