Automatic visual leakage inspection by using thermographic video and image analysis

Pipeline leakages are a critical issue in large-scale process plants, because leaks increase maintenance costs and create unsafe conditions. Therefore, detection of leakages is a crucial task for maintenance and condition monitoring. Recently, using IR cameras to detect leakages in large-scale plants was found to be a promising approach since IR cameras can capture images of leaking fluid if the fluid has a higher (or lower) temperature than its surroundings. In this paper, an approach based on thermographic data analysis using IR videos is proposed to detect leakage. In this approach, subsequent frames are subtracted to eliminate the background and reveal the motion of the leaking drops. Then, Principle Component Analysis is performed to extract features from the subtracted images. All subtracted images are individually transferred to feature vectors, which are considered as a basis for classifying the videos. Then, the K-Nearest Neighbor algorithm is used to classify the videos as normal (non-leakage) or anomalous (leakage). In order to evaluate the approach, a data set, consisting of video footage of a laboratory demonstrator plant captured by an IR camera, is considered. Leakages are simulated in the pipelines and the video data is used for image analysis. The results show that the proposed method is a promising approach to detect leakages from pipelines using IR video analysis.