Accelerometer-Based Motion Recognition and Alerting System for Abnormal Activity Detection in Dynamic Environments

Abstract

Anomaly Detection (AD) plays a crucial role in identifying patterns in data that deviate from expected behavior, especially in industrial environments where equipment failures can severely disrupt productivity. To address such challenges in real-time and resource-constrained settings, Tiny Machine Learning (TinyML) has emerged as a promising solution. This paper presents the design and implementation of a real-time activity recognition and abnormal motion detection system using accelerometer data and edge-based machine learning, specifically optimized through TinyML techniques. The system comprises eight modular components: motion sensing via a 3-axis accelerometer, continuous data acquisition, signal preprocessing, efficient inference through TensorFlow Lite, activity classification, command response control, alert generation using LEDs and buzzers, and a continuous monitoring loop. Motion signals are processed to extract statistical and temporal features, which serve as input to a pre-trained lightweight machine learning model designed for real-time inference on low-power embedded devices. The TinyML model, trained on a labeled dataset of human activity patterns, achieved a detection accuracy of 98.8% in distinguishing between normal and abnormal movements such as falls, sudden jolts, or prolonged inactivity. The model’s performance, measured through precision, recall, and F1-score metrics, indicates strong generalization and reliability for edge deployment. Abnormal activity triggers an immediate hardware response via audiovisual cues, facilitating timely interventions in sensitive environments such as elderly care, occupational safety, or remote health monitoring. Mathematical underpinnings of the system, including signal transformation, feature extraction, and binary classification logic, are explored in detail. The system operates entirely offline, requiring no external cloud connectivity, and supports continuous monitoring with minimal latency and energy consumption.