Key points are not available for this paper at this time.
Generally, the amount of wastewater in sewerage pipes is measured using sensor-based devices such as submerged area velocity flow meters or non-contact flow meters. However, these flow meters do not provide accurate measurements because of impurities, corrosion, and measurement instability due to high turbidity. However, cameras have advantages such as their low cost, easy service, and convenient operation compared to the sensors. Therefore, in this study, we examined the following three methods for measuring the flow rate by capturing images inside of a sewer pipe using a camera and analyzing the images to calculate the water level: direct visual inspection and recording, image processing, and deep learning. The MATLAB image processing toolbox was used for analysis. The image processing found the boundary line by adjusting the contrast of the image or removing noise; a network to find the boundary line between wastewater and sewer pipe was created after training the image segmentation results and placing them into three categories using deep learning. From the recognized water levels, geometrical features were used to identify the boundary lines, and flow velocities and flow rates were calculated from Manning’s equation. Using direct inspection and image-processing techniques, boundary lines in images were detected at rates of 12% and 53%, respectively. Although the deep-learning model required training, it demonstrated 100% water-level detection, thereby proving to be the most advantageous method. Moreover, there is enough potential to increase the accuracy of deep learning, and it can be a possible replacement for existing flow measurement sensors.
Ji et al. (Mon,) studied this question.