Robust Port Staff Trajectory Extraction under Adverse Visibility Challenges

Automated container terminals (ACT) are considered into the development direction of the port industry, and accurate movement data trajectories are essential to improve the safety of ACTs. Motivated by the new artificial intelligence techniques, we propose an integrated framework that extracts port operators in regions of interest from successive frames of port surveillance videos through a transformer-architecture codec and uses an improved depth-sorting algorithm to associate these port operator detection data with trajectories associated with extracting operator motion trajectories from low-visibility port surveillance video. The experimental results show that the multi-target tracking accuracy of operator trajectory extraction in the low-visibility port area is 99.66% and the multi-target tracking accuracy is 79.12%. In addition, the research results show that the framework can accurately extract operator trajectories in low-visibility port environments. It provides a technical guarantee for production safety in the complex automated port environment.