GaugeTracker: AI-Powered Cost-Effective Analog Gauge Monitoring System

featured.png

Abstract

Automating analog gauge readings is essential for providing stakeholders with timely alerts about abnormalities in physical properties measured by gauges, such as pressure, and for offering detailed historical data to improve understanding of the work environment. However, existing systems face challenges in balancing accuracy, continuity, reading latency, network bandwidth usage, and cost. In this study, we introduce GaugeTracker, an end-to-end system to address these challenges. Our proposed method, based on template matching for gauge reading, precisely determines the current angle of the gauge pointer, significantly outperforming state-of-the-art baselines with an average error of 1.81 degrees. By leveraging the versatility of large vision-language models, we develop a pipeline for automatically generating accurate and realistic gauge templates for each specific gauge at various readings on the server. Deployed on the world’s most affordable IoT camera, which is mounted in front of a gauge using our customized camera holder, our prototype system can read the gauge 7 times per second by processing entirely on the device. This delivers continuous and accurate gauge readings across diverse environmental conditions. Furthermore, with a cost of merely $10 per gauge, our system offers a highly cost-effective solution for real-time analog gauge monitoring.

Publication
In Proceedings of IEEE 7th International Conference on Multimedia Information Processing and Retrieval [MIPR2024]

Project Website: Click Here

Beitong Tian
Beitong Tian
Ph.D. Student in Computer Science

My research interests include wireless sensing network, mobile computing and machine learning.