A Real-Time Frame-Differencing CMOS Imager With ROI Filtering for Data and Power Reduction

This paper presents a computational image sensor design based on frame differencing to reduce output data volume, thereby improving system energy efficiency in Internet of Things (IoT) applications. The proposed design features an 8T-1C pixel with a dynamic source follower (DSF), enabling in-pixel frame differencing for motion detection without affecting image readout. The sensor integrates a multi-mode operation, offering full image output, data compression, and motion detection modes. A novel 10-bit two-step current-mode analog-to-digital converter (ADC), featuring a 3-bit flash and a 7-bit single-slope ADC, is introduced for digitalization. The 64×64 sensor prototype was implemented using a standard 0.18 μm CMOS process. Test results demonstrate the effectiveness of the DSF in enhancing frame differencing and reducing noise, leading to higher response linearity. Experimental validation of motion detection and data compression modes shows an average of 10.9% of output data volume compared to full frame, and over high reconstructed image quality. The results demonstrate that the proposed design achieves efficient, on-chip data reduction with low power consumption and enhanced adaptability for real-time applications.