Evaluating Electric Charge Variation Sensors for Camera-free Eye Tracking on Smart Glasses

TL;DR

This work addresses the problem of field evaluating a contactless QVar-based eye-tracking system for smart glasses. It combines a multi-channel QVar sensor setup with an on-device CNN classifier to assess real-world performance across 29 users and 100 recordings, including noise considerations. The study finds an average accuracy of 74.5% with substantial inter-subject variability (57–89%) and shows that ambient electromagnetic noise, notably from nearby laptops, degrades signal quality and classification. These findings demonstrate practical viability of contactless eye tracking in everyday settings while highlighting key challenges, guiding the development of more robust, adaptive gaze interfaces for wearable HCI and AR applications.

Abstract

Contactless Electrooculography (EOC) using electric charge variation (QVar) sensing has recently emerged as a promising eye-tracking technique for wearable devices. QVar enables low-power and unobtrusive interaction without requiring skin-contact electrodes. Previous work demonstrated that such systems can accurately classify eye movements using onboard TinyML under controlled laboratory conditions. However, the performance and robustness of contactless EOC in real-world scenarios, where environmental noise and user variability are significant, remain largely unexplored. In this paper, we present a field evaluation of a previously proposed QVar-based eye-tracking system, assessing its limitations in everyday usage contexts across 29 users and 100 recordings in everyday scenarios such as working in front of a laptop. Our results show that classification accuracy varies between 57% and 89% depending on the user, with an average of 74.5%, and degrades significantly in the presence of nearby electronic noise sources. These results show that contactless EOC remains viable under realistic conditions, though subject variability and environmental factors can significantly affect classification accuracy. The findings inform the future development of wearable gaze interfaces for human-computer interaction and augmented reality, supporting the transition of this technology from prototype to practice.

Figures (6)

• Figure 2: Accuracy on leave-one-subject-out cross-validation setup. sXX indicates the specific subject in the dataset left out during the training and only used for evaluation. This plot highlights the large intra-subject variability on classification accuracy.
• Figure 3: Median peak-to-peak value for each channel. Setup: non-contact: horizontal Qvar 0, vertical Qvar 1, horizontal corners Qvar 5; contact: diagonal left Qvar 2, diagonal right Qvar 3.
• Figure 4: T-SNE projections of Qvar data: (left) Subject 1 only, (middle) Subject 2 only, (right) Subjects 1 & 2 combined.
• Figure 5: Influence of a laptop on peak-to-peak noise picked-up by the 5 channels. Setup: non-contact: horizontal Qvar 0, vertical Qvar 1, horizontal corners Qvar 5; contact: diagonal left Qvar 2, diagonal right Qvar 3.
• Figure : (a) Electrode placement for the five differential channels in the glasses setup: non-contact: horizontal, vertical, horizontal corners; contact: diagonal left, diagonal right