In-Sensor Computing for Rapid Image Focusing during Liquid Lens Oscillation

Focus-tunable liquid lenses are a cornerstone in modern optical systems, enabling rapid and energy-efficient focus adjustments. However, the inherent fluidic dynamics of these lenses induce transient oscillations following focus adjustments, necessitating stabilization periods that introduce latency and hinder real-time autofocus performance. To overcome these limitations, this study presents a novel compact autofocus system leveraging an In-situ Vision Processor (IVP) coupled with a focus-tunable liquid lens. Our method introduces in-sensor parallel sharpness measurement and post-processed focused image acquisition during the liquid lens's oscillation phase directly on the sensor, eliminating the need for the liquid lens to stabilise before capturing images in most traditional liquid lens based vision systems. Consequently, the system is able to capture optimal in-focus images for both static and high-speed dynamic objects due to its high capture-process sampling speed. The ‘global focus’ image can also be generated for spatial objects while the lens sweeps between diopter extremes, enhancing the system's suitability for diverse applications. This technology operates with low power consumption (around 2.0 W), low latency (<2 ms for single image capturing and analysis), and low-cost ($ 10), making it suitable for mobile and portable applications.