Self Attention GAN and SWIN Transformer-Based Pothole Detection With Trust Region-Based LSM and Hough Line Transform for 2D to 3D Conversion
Vehicles and roads play a crucial role in connecting peoples across the world for their every day needs, activities and life style. This has enormously increased the need for maintaining the road surface with better quality and maintenance to prevent accidents and vehicles damages due to road hazard...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11108226/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Vehicles and roads play a crucial role in connecting peoples across the world for their every day needs, activities and life style. This has enormously increased the need for maintaining the road surface with better quality and maintenance to prevent accidents and vehicles damages due to road hazards such as cracks and potholes. In this paper, we have addressed the aforementioned problem with deep learning based General Adversarial Networks (GAN) along with Data Augmentation considering both road damaged and road damage free images from the datasets. Additionally, we have also employed SWIN transformer with Self-Attention Mechanism where 2D images are transformed into 3D images for better detection of potholes with improved accuracy using Trust Region Reflective Least Square Fitting Method and Hough line Transform along with Inverse Perspective Mapping is employed for dimensionality conversion of pothole dimensions. The proposed approach has considered modified YOLOv81 and the results were compared with basic YOLOv81 with various performance metrics. Our proposed approach has achieved 93.54% accuracy, minimum accuracy loss of 6.46% and highest F1-score of 76.42% while compared with ESRGAN-YOLOv5, ESRGAN-YOLOv7 and Potholes state of art approaches. |
|---|---|
| ISSN: | 2169-3536 |