Falls are a common cause of serious injury among elderly people, and the time until help arrives strongly affects the outcome. Inertial measurement units (IMUs) are already present in phones, smartwatches and cheap wearable boards, and provide a low-bandwidth, privacy-preserving alternative to camera-based monitoring. Simple threshold rules on acceleration magnitude confuse falls with sitting down quickly or dropping the device, so learned models are needed. Such a device runs on a microcontroller with limited memory and battery, so the model has to be small and fast enough to run continuously on-device.

Within this field, the following tasks can be performed:

• Literature Review & Technology Selection: Investigating state-of-the-art IMU-based fall detection methods, public datasets (e.g. SisFall, UP-Fall) and embedded machine learning toolchains.

• Data Processing Pipeline: Creating an automated pipeline to segment and window multi-axis inertial recordings and extract compact time- and frequency-domain features, together with augmentation to handle the small number of real fall events.

• Sequence Modeling: Designing or adapting a lightweight neural network architecture capable of classifying short temporal windows of IMU data (falls vs. activities of daily living) with a minimal parameter count.

• Model Optimization: Exploring and applying model compression techniques (such as quantization or pruning) to shrink the model footprint and prepare it for embedded deployment.

• Edge Deployment & Benchmarking: Deploying the end-to-end pipeline onto a target embedded edge device and analyzing the trade-offs between detection accuracy, false alarm rate, resource utilization, and real-time response time.

Requirements for the topic:

• Knowledge of Python programming

• Good English communication skills

Recommended for the topic:

• Basic knowledge of machine learning/deep learning concepts

• Minimal familiarity with Linux environments or single-board embedded systems

• Basic mathematical foundation (linear algebra and matrix operations)