CMOS Image Sensor with In-pixel Signal Processing for Fast Eye Tracking

Abstract

In this dissertation, an eye-tracking sensor integrated with in-pixel signal processing was presented. By embedding a detection unit directly within each pixel, the proposed sensor eliminated the need for image memory and off-chip interface circuits, thereby achieving both low power consumption and minimal processing latency. The detection unit was implemented entirely using digital logic and employed morphological image processing techniques—specifically shrinkage and expansion—to identify the center of the pupil or glint. This fully digital approach further contributed to power efficiency. To ensure robust performance under varying illumination conditions, an adaptive thresholding scheme was also developed. Both the binarization threshold voltage and photodiode integration time were dynamically adjusted to reduce latency without degrading detection accuracy. The proposed sensor, consisting of a 176 × 120-pixel array, was fabricated in a 180 nm standard CMOS process and experimentally validated. It demonstrated a localization error of approximately √2 pixels with a latency of 80 μs, while consuming only 5 mW from a 3.3 V power supply.