(in Japanese) Image Recognition of HDR Images with Dual Gain CMOS Image Sensor

2021 August

Authors Takuya Kitamura*, Kota Yoshida*, Tetsushi Ikeda*, Shinichi Ikenaga, Isao Takayanagi and Shunsuke Okura*
*Ritsumeikan University

映像情報メディア学会 情報センシング研究会

High dynamic range (HDR) CMOS image sensors are expected to be used for machine vision applications for recognition under extreme illumination conditions such as outdoors. A dual-gain readout image sensor has been proposed as one way to realize this HDR CMOS image sensor. However, to perform image recognition, it is necessary to convert 16-bit images captured by this image sensor into 8-bit images and apply them to a convolutional neural network (CNN) trained with 8-bit image data. In this paper, we report on the recognition results of images captured with an HDR CMOS image sensor under illumination environments with large contrast and various color temperatures.