Real-Time Panoramic Surveillance Video Stitching Method for Complex Industrial Environments
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.This study introduces a novel real-time video stitching method for industrial surveillance, enhancing feature extraction and optimizing seam lines to overcome challenging visual conditions and improve stitched video quality.
Area Of Science
- Computer Vision
- Image Processing
- Artificial Intelligence
Background
- Industrial surveillance videos often suffer from poor quality (low illumination, texture, overlap), hindering effective video stitching.
- Existing video stitching methods struggle with feature extraction and accurate registration in complex industrial environments.
Purpose Of The Study
- To develop a real-time panoramic video stitching method tailored for complex industrial surveillance scenarios.
- To enhance feature extraction and matching accuracy for improved image registration.
- To optimize the seam line selection to avoid moving objects and improve visual quality.
Main Methods
- Integrated Efficient Channel Attention (ECA) and Channel Attention (CA) modules with ResNet within the UDIS algorithm for enhanced feature extraction.
- Developed a novel loss function combining similarity loss (Lsim) and smoothness loss (Lsmooth) to refine registration.
- Introduced gradient and motion terms into the energy function for optimal seam line selection, enabling avoidance of moving objects.
Main Results
- The proposed registration method achieved superior performance with RMSE, PSNR, and SSIM values of 1.965, 25.338, and 0.8366, respectively.
- The fusion method resulted in smoother transitions and effectively avoided moving objects, significantly enhancing visual quality.
- The system achieved a real-time stitching frame rate of 23 fps, meeting industrial requirements.
Conclusions
- The proposed method significantly improves the accuracy and visual quality of panoramic video stitching in challenging industrial environments.
- The integration of attention mechanisms and a refined loss function enhances feature extraction and registration.
- The real-time performance makes the method suitable for practical industrial surveillance applications.
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
