earticle

영어

This study compares traffic object detection rates under various experimental conditions using YOLOv8. The mAP50 of the entire learning dataset was 0.952, indicating a stable detection rate that is expected to be put into practical use. In RGB CCTV images of an eight-lane academy street, the overall detection rate during the day was found to be higher than at night. Based on each road size, an eight-lane main road has a lower object detection rate than a two-lane side road, and pedestrians on the eight-lane road had lower predictive power than vehicles. This is believed to be due to the characteristics of walking in clusters at a crosswalk on main roads, simultaneously showing various and large amounts of mobility. From each image type, RGB showed higher predictive power than thermal images, and accordingly, RGB images are considered advantageous for overall traffic object control. It is hoped that follow-up research for the realization and practical use of technology will continue in the future, and through this, intelligent traffic control for public safety will be achieved.This study compares traffic object detection rates under various experimental conditions using YOLOv8. The mAP50 of the entire learning dataset was 0.952, indicating a stable detection rate that is expected to be put into practical use. In RGB CCTV images of an eight-lane academy street, the overall detection rate during the day was found to be higher than at night. Based on each road size, an eight-lane main road has a lower object detection rate than a two-lane side road, and pedestrians on the eight-lane road had lower predictive power than vehicles. This is believed to be due to the characteristics of walking in clusters at a crosswalk on main roads, simultaneously showing various and large amounts of mobility. From each image type, RGB showed higher predictive power than thermal images, and accordingly, RGB images are considered advantageous for overall traffic object control. It is hoped that follow-up research for the realization and practical use of technology will continue in the future, and through this, intelligent traffic control for public safety will be achieved.

한국어

본 연구는 교통사고 예방 목적의 CCTV 관제센터 기반 동적객체 감지 기술 실현 및 실용화 를 위한 기초연구로 YOLOv8을 활용하여 다양한 실험조건에서의 교통객체 감지율을 비교분석 하였다. 전체 학습 데이터셋의 mAP50은 0.952로 실용화가 기대되는 안정적인 감지율을 나타냈 다. 8차선 학원가사거리의 RGB CCTV 영상에서 주간의 전체 감지율이 야간보다 높게 분석되 었으며, 도로규모별 분석결과는 8차선 도로가 2차선 도로보다 객체 감지율이 떨어졌고, 8차선 도로에서 보행자 감지는 차량에 비해 예측력이 낮게 분석되었다. 학원가사거리 횡단보도에서 군집보행과 다량·다종의 모빌리티가 동시다발적으로 나타나는 특징 때문으로 판단된다. 영상 유형별 분석결과, 열화상보다 높은 예측력이 확보된 RGB 영상이 전반적인 교통객체 관제에서 적합한 것으로 나타났다. 향후 기술 실현 및 실용화를 위한 후속연구가 지속적으로 진행되어 국민안전을 위한 지능형 교통관제가 이루어지길 기대한다.