Automated Colilert detection for tap water
Author
Schmidt, Helge Thomsen
Term
4. term
Publication year
2018
Submitted on
2018-08-01
Abstract
Clean tap water is often taken for granted, yet slow contamination alerts can cause illness. This project explores whether image processing can automate Colilert-based monitoring of water samples to support an early warning system. A prototype was developed to handle the vision component: it captures images of Quanti-Tray/2000 samples throughout incubation, localizes the tray and wells using tray isolation, corner detection and perspective alignment, and analyzes well colors to assess growth. Detection thresholds were compared against a sample comparator, and module tests were conducted for tray and well localization and color analysis. The results indicate potential, and the system can follow a complete incubation phase, but performance was limited by the illumination setup, which led to inconsistent well localization. The work suggests that robust automation will require improved and more uniform lighting.
Rent drikkevand tages ofte for givet, men langsom varsling ved forurening kan føre til sygdom. Dette projekt undersøger, om billedbehandling kan automatisere Colilert-baseret overvågning af vandprøver for at støtte et tidligt varslingssystem. Der er udviklet en prototype, der håndterer den visuelle del af processen: Den optager billeder af Quanti-Tray/2000-prøver gennem inkubationen, lokaliserer bakken og brøndene ved hjælp af bakke-isolation, hjørnedetektion og perspektivjustering, og analyserer brøndfarver for at vurdere vækst. Systemets detektionstærskler blev sammenlignet med en reference-sample comparator, og der blev gennemført modultests for bakke- og brøndlokalisering samt farveanalyse. Resultaterne viser potentiale, og systemet kan følge en hel inkubationsfase, men ydeevnen blev begrænset af belysningsopsættet, som gav inkonsistent brøndlokalisering. Arbejdet peger på, at robust automatisering kræver forbedret og mere ensartet belysning.
[This apstract has been generated with the help of AI directly from the project full text]
Keywords