GazeTrak: Exploring Acoustic-based Eye Tracking on a Glass Frame (2402.14634v2)

Abstract: In this paper, we present GazeTrak, the first acoustic-based eye tracking system on glasses. Our system only needs one speaker and four microphones attached to each side of the glasses. These acoustic sensors capture the formations of the eyeballs and the surrounding areas by emitting encoded inaudible sound towards eyeballs and receiving the reflected signals. These reflected signals are further processed to calculate the echo profiles, which are fed to a customized deep learning pipeline to continuously infer the gaze position. In a user study with 20 participants, GazeTrak achieves an accuracy of 3.6{\deg} within the same remounting session and 4.9{\deg} across different sessions with a refreshing rate of 83.3 Hz and a power signature of 287.9 mW. Furthermore, we report the performance of our gaze tracking system fully implemented on an MCU with a low-power CNN accelerator (MAX78002). In this configuration, the system runs at up to 83.3 Hz and has a total power signature of 95.4 mW with a 30 Hz FPS.

• The paper introduces an innovative acoustic sensor approach that uses FMCW signals and deep learning to estimate gaze direction.
• It empirically evaluated the system with 20 participants, achieving 4.9° cross-session and 3.6° in-session tracking accuracy at 83.3 Hz.
• The design’s low power consumption, reducing usage to as little as 95.4 mW, highlights its potential for wearable real-time applications.

An Analysis of GazeTrak: Acoustic-based Eye Tracking for Smart Glasses

The paper "GazeTrak: Exploring Acoustic-based Eye Tracking on a Glass Frame" introduces a pioneering system for eye tracking using acoustic methods in a glasses form factor. Utilizing a custom setup of speakers and microphones, GazeTrak employs frequency-modulated continuous-wave (FMCW) acoustic signals to track eye movements. This setup stands out due to its low power consumption compared to traditional camera-based systems, making it particularly suitable for integration with devices like smart glasses, which often have limited battery capacity.

Key Approaches and Findings

1. Innovative Use of Acoustic Sensors: GazeTrak leverages the distortion of acoustic signals as they interact with the eyes and surrounding features to infer gaze direction. The system's hardware components include one speaker and four microphones per glass arm, utilizing encoded sound waves to generate an "echo profile." This profile is processed with a deep learning model to interpret gaze positions.
2. Deep Learning Integration: A specialized deep learning architecture based on the ResNet-18 framework was employed to handle the complex task of translating echo profiles into gaze estimations. The integration of machine learning allows for adapting to individual user characteristics and improving predictive accuracy over time.
3. Empirical Evaluation: The empirical section involved a paper with 20 participants, demonstrating that GazeTrak can achieve a cross-session tracking accuracy of 4.9 degrees and an even higher accuracy of 3.6 degrees within the same session. Such precision is achieved with a refresh rate of 83.3 Hz and modest power consumption, affirming the feasibility of real-time eye tracking on glasses.
4. Low Power Consumption: A significant achievement of GazeTrak is its low power signature — 287.9 mW on a Teensy 4.1 setup and down to 95.4 mW when fully implemented on a MAX78002 microcontroller with a custom low-power CNN accelerator. This represents a power reduction of over 95% compared to some existing camera-based solutions.
5. Robustness to Noise: The paper also confirmed GazeTrak's resilience under various environmental noise conditions, demonstrating its potential flexibility for diverse real-world applications without significant degradation of performance.

Practical Implications

The development of GazeTrak can be considered a substantial advancement in the domain of wearable eye-tracking technology. Its ability to operate effectively at low power levels increases the potential for prolonged active use without frequent recharging, something researchers and developers have long pursued in wearable devices. This makes it particularly suited to applications in mobile health, augmented reality, and assistive technologies.

Theoretical Implications and Future Work

Theoretically, GazeTrak encourages a shift in the paradigm of eye-tracking methodologies from camera-centric systems to acoustics-based ones. This transition leverages the innate benefits of acoustic methods, like low power requirements and unobtrusive setup, while presenting challenges such as maintaining precision across users and dynamic environmental conditions. Future research should focus on refining the model's accuracy further and reducing the amount of calibration data needed. Additionally, expanding the dataset for training the model and exploring diverse feature extraction methods could bolster generalizability to broader demographics without exhaustive personalization.

Conclusion

"GazeTrak: Exploring Acoustic-based Eye Tracking on a Glass Frame" demonstrates a promising alternative to traditional eye-tracking systems, pivoting on acoustic sensing's inherent advantages. Its utility in continuous, low-power, real-time tracking opens up various applications and sets a new benchmark for wearable technologies. As further innovations are made, GazeTrak and similar approaches could reshape the landscape of how gaze tracking is integrated into everyday technology.