딥러닝 기반 오염 영상 분류와 시맨틱 분할 모델을 활용한 자율주행 지속성 향상용 통합시스템
Integrated System with Deep Learning-Based Contaminated Image Classifier and Semantic Segmentation Model for Enhancing Autonomous Driving Persistence
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Cameras are one of the essential sensors in autonomous vehicles, but they are highly susceptible to contamination. Visibility degradation due to various weather conditions and contaminants can significantly impact the performance of vision-based driver assistance systems. Therefore, algorithms capable of quickly determining contamination status and assessing contamination levels are required. To address this, we propose an integrated system consisting of a Convolutional Neural Network-based image classifier, a semantic segmentation model, and a contamination decision algorithm. When processing a test dataset of 264 samples, the proposed integrated system reduced the false positive rate for contamination detection from 30.68% to less than 1% and achieved 30% improvement in processing speed. Experimental results on the NVIDIA Jetson AGX Xavier board show that the proposed system achieves a processing speed of over 30 frames per second, confirming its capability for real-time operation. The classification accuracy of the integrated system is 99.24%, and the IoU for opaque contamination in semantic segmentation is 0.8289, enabling the system to determine contamination status rapidly and accurately estimate contamination levels.
Abstract
Cameras are one of the essential sensors in autonomous vehicles, but they are highly susceptible to contamination. Visibility degradation due to various weather conditions and contaminants can significantly impact the performance of vision-based driver assistance systems. Therefore, algorithms capable of quickly determining contamination status and assessing contamination levels are required. To address this, we propose an integrated system consisting of a Convolutional Neural Network-based image classifier, a semantic segmentation model, and a contamination decision algorithm. When processing a test dataset of 264 samples, the proposed integrated system reduced the false positive rate for contamination detection from 30.68% to less than 1% and achieved 30% improvement in processing speed. Experimental results on the NVIDIA Jetson AGX Xavier board show that the proposed system achieves a processing speed of over 30 frames per second, confirming its capability for real-time operation. The classification accuracy of the integrated system is 99.24%, and the IoU for opaque contamination in semantic segmentation is 0.8289, enabling the system to determine contamination status rapidly and accurately estimate contamination levels.