PipeEyes: Contactless Pipe Fluid Speed Estimation With Millimeter-Wave Radar

Accurate pipe flow measurement is crucial for monitoring the operational status of pipeline systems and detecting abnormal leaks and blockages. However, most state-of-the-art solutions rely on intrusive physical sensors, requiring intrusive installation and high maintenance costs due to mechanical wear. Recent advances in wireless sensing technologies offer new opportunities for non-contact estimation. In this paper, we presentPipeEyes, a novel contactless pipe flow estimation system using Commercial-Off-The-Shelf millimeter-wave radar.PipeEyesexploits flow-induced spectral characteristics to estimate fluid speed. We develop a physics-guided model linking fluid speed to radar signals using dynamic pressure theory. To improve signal utilization, we propose a Multiple Channel Fusion algorithm that leverages wavelet denoising to integrate information across multiple radar channels. To tackle the weak and intricate spectrum induced by extensive noise, we develop a Mask-based Frequency Extraction method to extract the flow-related frequency components. Furthermore, motivated by the observation that harmonic components exhibit high similarity to the target fundamental frequency band in the spectrum, we design a Matching-based Augmentation algorithm to enhance the baseband spectrum using these harmonic components. To enable accurate fluid speed measurement, we extract representative spectral features and apply a custom lightweight network for speed estimation. We implement a prototype ofPipeEyeswith COTS mmWave radar, and extensive experiments conducted across diverse pipe types and flow conditions demonstrate an average estimation error of 12.6% over a fluid speed range of 0-30 m/min.