AbstractsComputer Science

A computer vision approach for sewer overflow monitoring

by Meng Han

Institution: University of Illinois – Urbana-Champaign
Department: 0106
Degree: MS
Year: 2015
Keywords: Computer vision
Record ID: 2060778
Full text PDF: http://hdl.handle.net/2142/72810


Combined Sewer Systems (CSS) exist in over 700 communities across the United States. Under extreme wet conditions, excess inflow which is beyond the capacity of CSS results in Combined Sewer Overflows (CSOs); the consequence being direct discharge of untreated water into the environment. Current CSO monitoring methods rely on in situ placement, where the sensors are installed within the combined sewer chambers and the harsh environment may decrease the expected lifetime of the sensors. Other limitations include high costs and accessibility difficulties for the sensing equipment. CSOs are a major concern for maintaining acceptable water quality standards and thus better monitoring is required. To overcome current CSO sensing limitations, this work has created a computer vision based approach for CSO monitoring from outlet points of CSS. This approach relies only on video capture of CSO events at outlet points where there is flow out of a CSS, thus a camera can be installed outside of the CSS without any contact with water. The proposed methodology is capable of detecting, identifying and tracking CSOs by motion, shape and color features. It is also able to measure flow rate based on a proposed model and two provided dimensions. Consequently, the approach can characterize CSOs in terms of occurrence, duration and flow rate. In addition, the algorithm package is implemented in a Windows desktop application for data visualization, and an iOS application for real-time CSO video capturing and processing. The computer vision approach was tested in a laboratory environment with three different flow rate conditions: 5, 15 and 25 gallons per minute. The performance was evaluated by comparing the results reported by the approach with the ground-truth baselines. The detection of an overflow event using the computer vision approach is 1.0 second slower than a ground-truth method. Flow rates reported by the computer vision approach are within 12\% from the ground-truth flow rate baseline. The results of this work have shown that computer vision can be used as a reliable method for monitoring overflows under laboratory conditions. It opens the possibility of applying computer vision techniques in CSO monitoring from outlet points with mobile devices in the field.