Smartphone-based Gaze Assessment by Eye Tracking
Authors: Holi, P., Dubey, V., Shearer, C., Thavanesan, K.
Conference: 15th IEEE International Conference on Image Processing Theory, Tools and Applications
Dates: 16/12/2026
Publication Date: 16/12/2026
Abstract:Horizontal gaze deviation is an important neurological sign evaluated in the National Institutes of Health Stroke Scale (NIHSS) during acute stroke assessments. Traditionally, this assessment is conducted manually by clinicians relying on visual observation, which can introduce subjectivity and variability, particularly in emergencies and telemedicine settings. This paper presents a lightweight approach for assessing horizontal gaze using a smartphone equipped with a standard front-facing RGB camera. The proposed system integrates real-time facial landmark detection, pupil centroid estimation, calibration-assisted gaze mapping, and time-series analysis of horizontal eye movements. A rule-based and clinically interpretable decision model classifies gaze behaviour into three NIHSS-based categories: normal gaze, partial gaze palsy, and forced gaze deviation. This approach was evaluated on 15 healthy participants across 45 controlled gaze-tracking trials under real-time mobile conditions. Experimental results showed stable pupil tracking and low-latency on-device inference using standard smartphone hardware. The proposed method achieved an overall classification accuracy of 91.1% (41/45 trials) while maintaining computational efficiency suitable for mobile deployment. However, this study represents a feasibility evaluation on healthy participants and does not constitute clinical diagnostic validation. Future work will focus on validation in stroke populations, robustness to unconstrained motion, and cross-device evaluation.
Source: Manual