Estimating water storage from images

Shahbaz, Ajmal; Yunas, Syed; Smith, Lyndon; Staddon, Chad

Abstract

This paper introduces a novel approach to estimate domestic water storage within households by leveraging the classical computer vision technique of object detection. Ensuring universal access to safe drinking water is a critical component of achieving the Sustainable Development Goals (SDG). In recent years, research priorities related to the SDG have evolved to encompass household-scale infrastructure and the real-world experiences of water insecurity. Climate change is dangerously affecting safe drinking water. While robust survey instruments have been developed for acquiring data on many crucial aspects of household water insecurity, such as the distance to water sources, the number of trips made, and experiences of water-related illnesses or injuries, methods for estimating household water storage still rely on manual inspection and estimation. Our proposed methodology involves the collection of a dataset from the Rohingya refugee camp, which is home to one million refugees. Initially, a subset of data is gathered from 900 households. This data is then meticulously cleaned and labeled with different classes, such as buckets, jugs, drums, and more, along with their respective storage volumes. Subsequently, the labeled dataset is used to train an object detection model, capable of identifying objects within images and precisely locating them. The detected objects are then associated with their respective storage containers, and their cumulative volumes are summed to provide a final estimated value within an image. We conducted experiments using five distinct object detection models, which yielded promising results.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 2023 IEEE International Conference on Big Data (BigData)
• Start Date: Dec 13, 2023
• End Date: Dec 18, 2023
• Acceptance Date: Nov 13, 2023
• Online Publication Date: Jan 22, 2024
• Publication Date: Jan 22, 2024
• Deposit Date: Jan 26, 2024
• Publicly Available Date: Jan 23, 2026
• Publisher: Institute of Electrical and Electronics Engineers (IEEE)
• Pages: 3375-3379
• Book Title: 2023 IEEE International Conference on Big Data (BigData)
• ISBN: 9798350324464
• DOI: https://doi.org/10.1109/BigData59044.2023.10386262
• Public URL: https://uwe-repository.worktribe.com/output/11628626
• Additional Information: Author list;; ; Ajmal Shahbaz ; Syed Yunas; Lyndon Smith ; Chad Staddon