This thesis addresses the challenge of protecting public IoT infrastructure against vandalism and theft. Traditional security measures, such as video surveillance, are too computationally and energetically demanding for autonomous, battery-powered operation in remote locations. As an alternative, we designed and implemented a reactive, low-power embedded system based on the ESP32 microcontroller and the LSM6DS3TR-C inertial sensor. We developed a multitasking software architecture within the FreeRTOS environment that, using hardware interrupts and the sensor's Continuous-to-FIFO operating mode, enables continuous sampling while preserving a critical 1-second pre-trigger time window. A complete preprocessing pipeline – including gravity offset removal via a high-pass IIR filter, magnitude calculation, and time-domain feature extraction – is executed in real time directly on the edge device. Final event classification is performed locally using an optimised decision tree. Evaluation on a field dataset of 810 samples showed that the system achieves 99.4\% classification accuracy and reliably distinguishes between physical attacks and wind-induced effects. By using a light-sleep mode, the device ensures several weeks of autonomous field operation, confirming the practical feasibility of low-power edge intelligence.
|