Junbo Zhao
Research Subjects
Research Interests
Junbo Zhao has a total of 1 co-authors. Here are the top co-authors by number of shared publications:
- Jinxiang Chen
College of Electric Power, Inner Mongolia University of Technology, Hohhot 010080, People’s Republic of China
1 shared publication
Latest Publications
- Open Access
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 409-415
YOLOv8 Detection and Improved BOT-SORT Tracking Algorithm for Iron Ladles
Iron ladles play a significant role in the industrial intelligence upgrade of steel plants. Accurate recognition and tracking for moving iron ladles can provide the location, speed, and operations information of iron ladles, which are essential for making scheduling plans for steel production. YOLOv8 detection and state-of-the-art (SOTA) tracking algorithms for iron ladles are presented in this paper. The Video data sets with or without shelters are constructed by collecting the actual iron ladles working data. Some own image and video datasets are added to the above datasets by using Segment Anything (SAM) and DarkLabel due to lack of iron ladles data. The YOLOv8 detection model is applied to detect the iron ladles, and three trackers, which are the StrongSORT, OC-SORT, and BOT-SORT, are applied to achieve real-time position information of iron ladles, respectively. In order to improve the identification and tracking accuracy for iron ladles, a genetic algorithm is used to optimize the parameter of the above three trackers. The training and testing results of the above model show that the BOT-SORT tracking model with genetic optimization achieves the highest accuracy that HOTA score is 97.49, both MOTA and IDF1 are 100.
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 409-415
Open AccessYOLOv8 Detection and Improved BOT-SORT Tracking Algorithm for Iron Ladles
Iron ladles play a significant role in the industrial intelligence upgrade of steel plants. Accurate recognition and tracking for moving iron ladles can provide the location, speed, and operations information of iron ladles, which are essential for making scheduling plans for steel production. YOLOv8 detection and state-of-the-art (SOTA) tracking algorithms for iron ladles are presented in this paper. The Video data sets with or without shelters are constructed by collecting the actual iron ladles working data. Some own image and video datasets are added to the above datasets by using Segment Anything (SAM) and DarkLabel due to lack of iron ladles data. The YOLOv8 detection model is applied to detect the iron ladles, and three trackers, which are the StrongSORT, OC-SORT, and BOT-SORT, are applied to achieve real-time position information of iron ladles, respectively. In order to improve the identification and tracking accuracy for iron ladles, a genetic algorithm is used to optimize the parameter of the above three trackers. The training and testing results of the above model show that the BOT-SORT tracking model with genetic optimization achieves the highest accuracy that HOTA score is 97.49, both MOTA and IDF1 are 100.