| 제목 | 횡단보도 보행자 인지 향상을 위한 ByteTrack 기반 다중객체추적 |
|---|---|
| Title | ByteTrack-Based Multi-Object Tracking for Recognition of Pedestrian in Crosswalk |
| 저자 |
노유진 (광운대학교 소프트웨어융합대학) 이유진 (광운대학교 소프트웨어융합대학) 이상민* (광운대학교 소프트웨어융합대학) |
| Author |
Yujin Roh(College of Software and Convergence, Kwangwoon University) Yujin Lee(College of Software and Convergence, Kwangwoon University) Sangmin Lee†(College of Software and Convergence, Kwangwoon University) |
| Bibliography | Journal of Logistics Science & Technology, 3(2),56~77, 2022, |
| DOI | |
| Key Words | Pedestrian Recognition, Multi-Object Tracking, Autonomous Vehicle, ByteTrack, CDNet |
| Abstract | According to traffic accident statistics, 42% of traffic accidents between 2017 and 2021 were crossing accidents. To reduce these accidents, the research on the collision avoidance assistance for pedestrian safety is in the spotlight in machine learning field. However, there are few studies to improve the accuracy of recognizing the crossing multiple pedestrians and estimating the residual time to the crossing. We here propose using hybrid approach combining Yolo-X and ByteTrack to obtain the accurate detector of pedestrians. We gathered the actual data set of the pedestrian crossing for the accurate multi-pedestrian tracking model. The proposed approach can detect and track the crossing pedestrians in real time. Furthermore, we can estimate the pedestrian’s walking direction, speed, acceleration, and distance to the crosswalk for inferring the motion of pedestrians entering the crosswalk. The experimental results present that the proposed approach outperforms the alternatives in terms of tracking accuracy and computation cost. |
| PDF download | JM_3[2]-P56~77.pdf |