研究目的
To design a CMOS image sensor for high-speed and low-latency eye tracking, specifically for real-time saccade tracking.
研究成果
The designed CMOS image sensor successfully achieves high-speed (315 fps) and low-latency (single frame) eye tracking, enabling real-time detection of saccade motions with angular velocities up to 250 deg/s. This demonstrates the effectiveness of the column-parallel architecture for improving eye tracking systems, with potential applications in user interfaces. Future work could focus on automating ROI configuration and enhancing accuracy.
研究不足
The study has limitations such as manual configuration of region of interest (ROI) and threshold voltage adjustment for pupil digitization, which could introduce errors; presence of black regions other than the pupil may affect centroid calculation accuracy; the experimental setup requires manual calibration and is not fully automated; the sensor's performance might be constrained by the specific CMOS process used (0.18 μm).
1：Experimental Design and Method Selection:
The study employs a column-parallel processing architecture in a CMOS image sensor to achieve high-speed calculations of pupil centroid (S, SX, SY) for line-of-sight detection, using a centroid calculation algorithm based on binary pupil area flags.
2：Sample Selection and Data Sources:
The experiment uses a fabricated CMOS image sensor with 640x480 pixels, and human subjects for eye tracking tasks, such as reading sentences.
3：List of Experimental Equipment and Materials:
Includes the fabricated CMOS image sensor (
4：18 μm CIS process), evaluation board with FPGA for control signals, A/D converter for raw image capture, IR-LEDs for illumination, chin-rest for subject positioning, and PC for data transfer and analysis. Experimental Procedures and Operational Workflow:
The sensor captures infrared eye images at 315 fps with 980 μs exposure time; pupil position is calculated in real-time using the column-parallel architecture; data is transferred to PC for recording and analysis of eye motions and angular velocities.
5：Data Analysis Methods:
Pupil centroid is calculated from S, SX, SY values; angular velocity is derived using a model based on eye ball radius and pixel size; statistical analysis includes averaging over successive samples to identify saccade motions.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
