INTELLIGENT PIPELINE DEPOSIT TRACKING BASED ON A MULTI-OBJECT TRACKING FRAMEWORK
Seminars
Semester 2
Pipelines play a significant role in transferring energies, materials and fulfilling public needs. However, conventional pipeline maintenance approaches predominantly depend on human inspection of captured closed circuit television (CCTV) records, a process that is particularly labor-intensive and time-consuming for lengthy pipelines. To address these limitations, this study proposes an artificial intelligence-based autonomous framework based on the multi-object tracking (MOT) algorithm for efficient and accurate deposit detection and tracking within pipelines, significantly reducing the need for manual intervention. The proposed MOT model has been trained and validated on a customized pipe CCTV dataset, consisting of more than 12,000 video frames. The experimental results indicate that the combination of YOLOX (for detection) and BYTE (for tracking) achieves the best overall performance among all the tested models. Further testing conducted on a real-world sewer pipeline project demonstrates the robustness of our model. The estimation error of the deposit location predicted by the MOT model is less than ± 0.1 m, with a mean absolute error of only 0.06 m. These findings demonstrate that the proposed autonomous MOT system offers clear advantages over manual inspection, including higher efficiency, reliable accuracy and reduced labor demands, possessing a strong potential for practical engineering applications.
For additional information, please contact Mr. Sihao YU, sihaoyu@connect.hku.hk.