基于人眼追踪的手机解锁系统 报告人：李映辉 指导老师：王继良 Unlock with A Glance 基于人眼追踪的手机解锁系统 报告人：李映辉 指导老师：王继良
Targets Gaze-gesture based unlocking system Gaze tracking implementation Accuracy of under 3° Tracking speed of more than 10 fps Allow free head motion Gesture recognition method Gesture is easy to perform by human eyes Robust to bias and error of gaze estimation Ability to recognize directional lines
Transfer into Directions Design Gaze Estimation Transfer into Directions Record Path Authentication
Challenges Sensitivity to eye localization accuracy At a distance of 30cm, 1 pixel corresponds to 0.169mm Fine grained eye localization is time-consuming (50% processing time) Robust gesture recognition
Observation Both head and eyes move when people look at different directions in a natural way Head movement arise more changes on eyes’ positions, which decreases the sensitivity of eye localization Distribution of left eye position when head motion is forbidden Distribution of left eye position when head motion is enabled
Gaze Tracking History Data kNN Search Gaze Position Calibration Parameter Estimation Touch Position Standardized Eye Position RGB image Eye Detection
Gesture Recognition → ↙ Direction Bucket Direction Calculator (robust fitting) Slide Window Gaze Trace List
Evaluation Gaze Tracking Performance Gaze gesture recognition evaluation Method Accuracy(°) Speed(fps) Gazture (2.6, 1.2) 13.8 EyeTab 6.88±1.8 12 GT on Tablet 4.42 0.7 Pupil-Canthi-Ratio 3.9 NA P1 P2 P3 P4 Z 1 2 3 4 L 8 N 10
Finished Work Gaze tracking methodology with accuracy of under 3° and speed of more than 10 fps Gaze gesture recognition method Evaluation of the usability of the system
Timeline 周次 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 工作安排 文献阅读 方案设计 眼球追踪算法实现 解锁功能实现 中期答辩 后期性能优化 撰写论文 期末答辩
THANKS FOR LISTENNING Q & A