MicroNAS: An Automated Framework for Developing a Fall Detection System

TL;DR

The paper addresses the need for real-time fall-detection on memory-constrained microcontrollers by introducing MicroNAS, an MCU-aware neural architecture search framework that uses the ESP32's memory as a deployment constraint. It demonstrates that single-stage, memory-guided NAS can yield deployable time-series classifiers (1D CNN and GRU) that outperform ensemble methods and a general AutoML baseline, while achieving substantially smaller memory footprints than pruning-based NAS. The pilot fall-detection study with 35 participants shows MicroNAS models achieve higher F1-scores than baselines, indicating practical viability for body-worn IMU-based systems in prosthetic contexts. The work provides open-source tooling and a methodology for memory-aware model design, enabling biomechanists to tailor FDS to microcontroller platforms and move toward real-world deployment, albeit with acknowledged limitations in dataset size and lab-only validation.

Abstract

This work presents MicroNAS, an automated neural architecture search tool specifically designed to create models optimized for microcontrollers with small memory resources. The ESP32 microcontroller, with 320 KB of memory, is used as the target platform. The artificial intelligence contribution lies in a novel method for optimizing convolutional neural network and gated recurrent unit architectures by considering the memory size of the target microcontroller as a guide. A comparison is made between memory-driven model optimization and traditional two-stage methods, which use pruning, to show the effectiveness of the proposed framework. To demonstrate the engineering application of MicroNAS, a fall detection system (FDS) for lower-limb amputees is developed as a pilot study. A critical challenge in fall detection studies, class imbalance in the dataset, is addressed. The results show that MicroNAS models achieved higher F1-scores than alternative approaches, such as ensemble methods and H2O Automated Machine Learning, presenting a significant step forward in real-time FDS development. Biomechanists using body-worn sensors for activity detection can adopt the open-source code to design machine learning models tailored for microcontroller platforms with limited memory.

Figures (7)

• Figure 1: IMU sensors on both shanks of lower limb amputee, red circles indicate IMU locations.
• Figure 2: Data division into Training, Validation, and Test sets. The Test set consisted of a single participant with a lower limb amputation (Combination 1) or no lower limb amputation (Combination 2). Each participant was in the Test set once creating 35 unique test data sets for evaluating model performance and statistical testing.
• Figure 3: Windowing with overlap. T indicates the time stamp where T1 is the sample recorded at 0.01 seconds and T2 at 0.02 seconds during 100 Hz recording (Figure does not show time windows to scale).
• Figure 4: Model development and stages in MicroNAS
• Figure 6: Windowing on Ensemble model predictions. Each prediction is for each time step and majority voting is used on 120 consecutive predictions.