An Energy-efficient Adaptive CMOS Image Sensor.

[摘要] In this thesis, an energy-efficient CMOS image sensor has been studied to adaptively provide low-power consumption, high dynamic range, and reduced spatial-temporal bandwidth for applications in distributed sensor networks and wireless biomedical imaging systems. In wireless sensor nodes, imagers typically have four main challenges to address: low power operation from limited energy sources, high dynamic range imaging to cover a wide range of illumination, high spatial resolution for high-quality imaging, and high temporal resolution for video streaming. However, it is difficult to optimize all of these as there are trade-offs among these four parameters and additionally the environmental conditions continuously change, requiring optimization of sensor operation on-the-fly. One of the solutions to this trade-off is implementing an imager with situation-aware adaptability. In this work, three on-chip adaptive functions have been investigated including (1) energy-adaptive imaging for low-power operation, (2) illumination-adaptive imaging for high-dynamic range, and (3) object-adaptive imaging for temporal-spatial bandwidth suppression. For proofs of concepts of this adaptive imaging, we designed two prototype chips. In the first chip, we implemented the energy/illumination adaptation by reconfiguring the mode of operation. Most of the time, the sensor operates in a monitoring mode at an extremely low power (less than 1.36 μW/frame). It switches to either a high-sensitivity or a wide-dynamic range mode of operation depending on illumination conditions. This illumination adaptation provides reconfigurability for sensitivity and dynamic range. The sensor can provide a high sensitivity of 23.9 V/lx∙s at low illumination, but it can be reconfigured to generate a high dynamic range of up to 99 dB in high illumination on-the-fly. In the second chip, we implemented motion-triggered, object-adaptive imaging to suppress the redundancies in spatial-temporal bandwidths in image signals. The sensor will wake up when triggered by motion and extracts features from the captured image for the detection of objects-of-interest. Full image capture operation is performed only when the objects-of-interest are found, resulting in a significant reduction of total power consumption at the sensor node. Object detection has been performed using the classification algorithm and has achieved a 94.5 % success rate for human recognition from 200 test images.

[发布机构] University of Michigan

[效力级别] High Dynamic Range

[关键词] An Energy-efficient CMOS Image Sensor Adaptively Provides Low-power Consumption;High Dynamic Range;and Reduced Spatial-temporal Bandwidth;Electrical Engineering;Engineering;Electrical Engineering