Liquid leaks are rare, but when they occur they can pose a significant security problem, depending on the escaping medium. Additionally, it requires clean-up work and may lead to delays in production or supply. At the moment, such detections are handled by sensors in or on the pipe. However, leaks are not always registered quickly enough by them, especially if the leaking medium happens only in small quantities. The thesis examines the feasibility of a visual detection approach for liquid leaks by evaluating various algorithms from the Computer Vision library OpenCV.

In a first section of the work, various methods for the detection of a scene change are examined by means of a literature review. This serves as the foundation for the experimental phase, in which video recordings of self-defined test cases are fed into algorithms in order to assess the success rate of detection of liquid leaks. All algorithms are based on the principle of Computer Vision and are evaluated in the respective test cases and compared with each other.

The experiment has proven the possibility of detecting liquid leaks in an industrial environment based on a Computer Vision approach. However, it turned out that visual recognition also poses challenges, with the influence of lighting being just one example that can certainly be optimized. However, the documentation of existing methods as well as the knowledge gained from practical application serve F. Hoffmann-La Roche Ltd as the basis for further development not only with regard to the detection of liquid leaks, but also other areas of application with the requirement for visual detection.