EVM-Insight: A Low-Cost Eulerian Video Magnification System for Structural Health Monitoring

EVM-Insight is a low-cost, open-source structural health monitoring (SHM) system that combines Eulerian Video Magnification (EVM) with inertial ground-truth validation and AI-based crack detection into a unified, field-deployable inspection pipeline. The system amplifies sub-millimetric structural vibrations up to ×100 using a Laplace pyramid decomposition and temporal Butterworth filtering applied to video captured by a fixed camera (DJI Osmo Pocket 3 or Nikon D7200 with telephoto lens). Optical frequency estimates are cross-validated against simultaneous accelerometer measurements (Sense HAT LSM9DS1 / MPU-6050), with a target agreement error below 15%. A secondary pipeline runs YOLOv8-nano on the POCO X7 Pro neural processing unit (NPU) for real-time crack detection, spatially correlated with the EVM vibration map to produce a zone-level risk index and an automated PDF inspection report — entirely in the field, without a laptop. The entire system is built on hardware already owned by the author (Raspberry Pi 4, ESP32, Arduino Mega 2560, consumer cameras) with zero software licensing cost. The Raspberry Pi 4 operates as a WiFi hotspot and EVM processing engine; the POCO X7 Pro serves as the AI inference engine and live dashboard terminal. This is a pre-experimental technical report presenting the system architecture, five-layer field topology, EVM processing pipeline, ground-truth validation methodology, ten-phase implementation roadmap, and the scientific rationale for each design decision. No experimental results are reported. Field validation results will be published in a subsequent version upon completion of Phase 5.