Lock Water Level Estimation Using CCTV Imagery

In 2022, 480 million tons of cargo valued at over $186 billion transited U.S. inland waterways through locks maintained by the U.S. Army Corps of Engineers (USACE). USACE uses the Lock Performance Monitoring System (LPMS) to capture data such as water level and lockage configuration. LPMS data are manually entered by staff, who are also responsible for lock operation and ensuring safe transit. This manual data entry process, extraneous to actual lock operation, adds to the cognitive load of operators during safety-critical times. This study is part of a larger effort to evaluate using close circuit video imagery (i.e., CCTV) to automate LPMS entry, starting with water level. For proof of concept, researchers calibrated one CCTV camera at McAlpine Lock and Dam in Louisville, KY and performed water level extraction on 1,131 quarter-hourly images captured 24/7 over a 2-week period. Water levels were detected via the zero-shot Segment Anything Model and projected into geospatial elevations via photogrammetric camera calibration. The results compared well with up/downstream in situ gauges, with a root mean squared error of 7.3 cm after a 15.9-cm constant bias and outliers were removed, comparable to wall staff gauge precision. Future steps include automating outlier (vessel obstruction) removal via vessel detection, and estimating flotilla size, draft, and configuration once detected. This study serves as a successful proof of concept for further operational development. Automated LPMS entry via CCTV has the potential to increase staff efficacy, lockage safety, and provide more accurate data for lock operation optimization across USACE.