关键词: CMOS图像传感器, 读出电路, 低功耗, 卷积神经网络, 人脸检测

Abstract: In an image sensor used for face detection, both frame rate and power consumption are key metrics. To address the issues of low frame rate and high power consumption in intelligent face detection image sensors, a high-speed and low-power readout circuit is proposed. The convolutional neural network (CNN) is used for face detection, which compresses the output data of the original image by 66.6% and reduces the number of analog-to-digital conversions. The CNN used can achieve 3×3 convolutions, rectified linear units, 2×2 max pooling, and 1×1 fully connected layers. The readout circuit supports configurable timing, making the readout circuit applicable in both grayscale image mode and CNN mode through different timing configurations. The circuit is designed based on 55-nm CMOS process, with a supply voltage of 1.2 V. In the CNN mode, the output frame rate reaches 603 frame/s, and the total power consumption of the readout circuit is 137 μW. The accuracy of face detection achieves 98.3% in the Labeled Faces in the Wild dataset.

Key words: CMOS image sensor, readout circuit, low power, convolutional neural network, face detection