모델 기반 이미지 디-스노잉 알고리즘 개발
Model-based image de-snowing algorithm development
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문은 자율주행기술에 활용 가능한 디-스노우 알고리즘 개발을 목표로 한다. 자율주행기술에 디-스노우 알고리즘을 적용하기 위해서는 속도와 성능을 보장 할 수 있어야 한다. 이를 위해, 노이즈 제거 알고리즘의 연산속도와 다양한 접근방식에 대해 분석하고, 모델 기반의 이미지 디-스노잉 알고리즘을 제안하였다. GPU 병렬연산이 가능하도록 알고리즘을 구축하여 기존 노이즈 제거 알고리즘 대비 속도를 대폭 향상시켰으며, 대상 픽셀의 노이즈 여부를 판정할 수 있는 햄펄 필터를 적용하여 이미지 품질을 보장하면서 노이즈 제거 성능을 발휘한다.
Abstract
This paper was written to develop a de-snowing algorithm that can be used for autonomous driving technology in a snowy environment. To apply the de-snowing algorithm to autonomous driving technology, it is necessary to ensure a fast computation speed. Also, it should be possible to visually remove snow particles while preserving image quality. To implement this, we analyzed the computation speed of the state-of-the-art noise removal algorithms in consideration of the computation speed required in the autonomous driving object detection research field. Besides, considering the approach of the noise removal algorithm, an approach that can guarantee noise removal performance under various environmental conditions was set as a target. As a result, a model-driven image de-snowing algorithm was proposed. We designed the algorithm to take advantage of the GPU hardware power in terms of speed. Concerning accuracy, a filter of the Hampel concept that can determine whether the target pixel is noisy was applied. Through a filter, pixels that are not noise are preserved to ensure image quality, and noise can be removed by replacing noise pixels from a reference frame.
참고문헌 (7)
- Li, Minghan, "Video rain streak removal by multiscale convolutional sparse coding", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2018)
- Jiang, Tai-Xiang, "Fastderain : A novel video rain streak removal method using directional gradient priors", IEEE Transactions on Image Processing, 28(4), pp.2089~2102 (2018)
- Jiang, Tai-Xiang, "A novel tensor-based video rain streaks removal approach via utilizing discriminatively intrinsic priors", the ieee conference on computer vision and pattern recognition (2017)
- Yang, Wenhan, "Self-Learning Video Rain Streak Removal: When Cyclic Consistency Meets Temporal Correspondence", Proceedings of the IEEE, 2020, ~
- Bochkovskiy, Alexey, "YOLOv4: Optimal Speed and Accuracy of Object Detection", ~ (2020)
- Wang, Hong, "A survey on rain removal from video and single image", ~ (2019)
- Garg, Kshitiz, "Vision and rain", International Journal of Computer Vision, 75(1), pp.3~27 (2007)