Eye movement analysis for activity recognition using electrooculography.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Signal Processing

Background:

• Activity recognition traditionally relies on sensors like accelerometers or cameras.
• These conventional methods have limitations in detecting subtle or complex human activities.
• Eye movement analysis offers a novel, non-intrusive sensing modality.

Purpose of the Study:

• To investigate eye movement analysis via electrooculography (EOG) as a new sensing modality for activity recognition.
• To develop and evaluate algorithms for detecting eye movement characteristics and repetitive patterns.
• To assess the feasibility of person-independent eye-based activity recognition (EAR).

Main Methods:

• Eye movement data were collected using an electrooculography (EOG) system.
• Algorithms were developed to detect saccades, fixations, and blinks from EOG signals.
• A feature set of 90 characteristics was engineered, with a subset selected using minimum redundancy maximum relevance (mRMR).
• A support vector machine (SVM) classifier was employed for person-independent (leave-one-person-out) training and validation.

Main Results:

• The study involved eight participants in an office environment, recognizing five activities plus a NULL class.
• The proposed method achieved an average precision of 76.1% and recall of 70.5% across all participants and activities.
• The system demonstrated effectiveness in distinguishing between activities like reading, note-taking, and web browsing.

Conclusions:

• Eye movement analysis (EAR) is a promising new sensing modality for activity recognition.
• The developed EOG-based method shows potential for recognizing activities difficult for conventional sensors.
• Further research can explore the broader applicability of EAR in diverse human-computer interaction scenarios.