# Open-Set Recognition of Human Activities from Head-Mounted Inertial Sensor

## TL;DR

This paper shows how to detect unknown activities using lightweight methods with head-mounted sensors, without changing the model or retraining.

## Contribution

A lightweight open-set recognition framework for HAR is proposed, compatible with wearable platforms and requiring no model retraining.

## Key findings

- Low-complexity open-set methods can detect unknown activities from head-mounted inertial data.
- Open-set HAR achieved robust performance with area under the ROC curve > 0.8.
- The methods operate in the logit space with minimal computational overhead.

## Abstract

What are the main findings?
Existing head-mounted human activity recognition (HAR) classifiers can be extended to detect previously unseen activities by adding lightweight, post hoc open-set mechanisms without modifying the model architecture or retraining procedure.A rigorous leave-one-activity-out evaluation shows that effective unknown activity detection can be achieved from head-mounted inertial data using low-complexity open-set scoring methods.

Existing head-mounted human activity recognition (HAR) classifiers can be extended to detect previously unseen activities by adding lightweight, post hoc open-set mechanisms without modifying the model architecture or retraining procedure.

A rigorous leave-one-activity-out evaluation shows that effective unknown activity detection can be achieved from head-mounted inertial data using low-complexity open-set scoring methods.

What are the implications of the main findings?
The lightweight nature of the explored techniques indicates their potential compatibility with resource-constrained wearable platforms.The empirical findings provide practical insights for designing real-world open-set HAR systems, particularly regarding dataset composition.

The lightweight nature of the explored techniques indicates their potential compatibility with resource-constrained wearable platforms.

The empirical findings provide practical insights for designing real-world open-set HAR systems, particularly regarding dataset composition.

Human activity recognition (HAR) based on inertial measurement units (IMUs) embedded in wearable devices has gained increasing relevance in healthcare, wellness, and fitness monitoring. However, most existing classification methods assume a closed-set setting, where all activity classes need to be defined during training, which limits their applicability in real-world environments where unseen or unexpected activities are present. To overcome this limitation, we adopt an open-set recognition (OSR) framework that requires minimal changes to the HAR classifiers traditionally employed for this purpose. We also provide an extensive empirical evaluation based on a leave-one-activity-out validation protocol applied to two datasets with IMU signals acquired from smart eyewear: a proprietary dataset and the publicly available UCA-EHAR dataset. A lightweight one-dimensional convolutional neural network was trained to classify six-axis IMU data across common activities. We assess open-set HAR performance using several methods requiring limited computational overhead and operating in the logit space, including maximum logit, Gaussian Mixture Models, Kernel Density Estimation, OpenMax, and Nearest Neighbor Distance Ratio. Robust identification of unknown activities was achieved, with area under the ROC curve > 0.8. These findings highlight the potential of low-complexity open-set approaches for real-time HAR on resource-constrained wearable platforms, supporting the development of adaptive and reliable sensor-based recognition systems for real-world use.