Turbidity measurement based on computer vision
Author
Hansen, Lærke Isabella Nørregaard
Term
4. term
Publication year
2019
Abstract
Stationary underwater cameras capture large volumes of video, but image quality is often degraded by turbidity, i.e., light scattered and absorbed by suspended particles. This thesis investigates whether turbidity can be measured directly from underwater video using computer vision, and how such measurements can be applied. The work includes a controlled experiment in a water tank where turbidity is varied (including different particle types) under defined lighting and camera settings. From these data, turbidity-related image features are extracted, data are normalized, and a model is developed to estimate turbidity. The resulting insights are then transferred to a real-world setup with cameras at Limfjordsbroen, where videos are imported, preprocessed, and annotated, and influence factors such as currents, weather, and traffic are considered. The thesis reports results from both laboratory and field data and discusses how turbidity measures can explain periods of poor image quality, support monitoring and planning, and inform future work, including measurement without reference instruments, motion detection, and correlation with weather conditions.
Under vand optager stationary kameraer store mængder video, men billedkvaliteten forringes ofte af turbiditet, dvs. lys, der spredes og absorberes af suspenderede partikler. Dette speciale undersøger, om turbiditet kan måles direkte fra undervandsvideo ved hjælp af computer vision, og hvordan sådanne mål kan anvendes. Arbejdet omfatter et kontrolleret eksperiment, hvor vandets turbiditet varieres (bl.a. med forskellige partikeltyper) i en vandtank med defineret lys- og kamerasetup. Herfra udtrækkes turbiditetsrelaterede billedfeatures, data normaliseres, og der udvikles en model til estimering af turbiditet. Den opnåede viden overføres derefter til et real-world setup med kameraer ved Limfjordsbroen, hvor videodata importeres, forbehandles og annoteres, og indflydelsesfaktorer som strøm, vejr og trafik diskuteres. Specialet præsenterer resultater fra både laboratorie- og feltdata og diskuterer, hvordan turbiditetsmål kan bruges til at forklare perioder med dårlig billedkvalitet, understøtte overvågning og planlægning, samt pege mod muligheder for fremtidig arbejde, herunder måling uden referencer, bevægelsesdetektion og korrelation med vejrudvikling.
[This apstract has been generated with the help of AI directly from the project full text]