Using computer vision to monitor ice conditions in water supply infrastructure: a study of salient image features

Abstract

Ice condition monitoring (ICM) is critical for the operation and maintenance of water supply infrastructure in cold regions. Existing approaches either depend on ground-level sensors or satellite photography for ICM, which suffer from high maintenance costs or inadequate precision. Computer vision (CV) has the potential to tackle the limitations by providing a precise and scalable solution based on near-shore cameras and increasingly affordable drones. To explore the potential of CV for ICM, this paper presents a systematic study of salient image features for differentiating typical ice evolvement phases throughout the freeze–thaw cycle. First, ice condition during the freeze–thaw cycle is studied to provide a categoric system of typical ice stages. Second, multiple image feature descriptors are proposed to characterize the distinction between different ice conditions. Finally, with the proposed descriptors as input, two support vector machines (SVMs) are trained to classify the ice condition for automatic ICM. Experiments have been implemented to identify salient features for ice characterization. It was found that the SVMs can achieve 71.9 and 77.3% accuracy for the prediction of ice stage and ice flow strength, respectively. Future research is suggested to develop the research findings into practical solutions for webcams or drone-based automatic ICM. HIGHLIGHTS Computer vision is used to monitor ice conditions in water supply infrastructure.; Image features are handcrafted to characterize different ice conditions.; Effectiveness of the features is quantified and evaluated by correlation analysis.;

Keywords

• feature engineering
• ice condition
• image recognition
• infrastructure
• machine learning
• water channels