Single-Shot Efficient Depth Imaging Based on Time-Compressive CMOS Image Sensor

概要

This article introduces a novel framework that effectively resolves multipath interference (MPI) in time-of-flight (ToF) depth measurements by utilizing a single-shot time-compressive complementary metal–oxide– semiconductor (CMOS) image sensor, a U-Net-based neural network, and joint end-to-end optimization. Conventional indirect ToF (iToF) methods inevitably suffer from MPI artifacts because each detection tap integrates incoming photons over a wide exposure window, making it difficult to isolate the direct reflection component from multiple reflections or scattering. Our approach, termed coded ToF (cToF), builds upon the iToF principle but employs coded exposure patterns to capture the scene response in a single shot. This time-compressive acquisition allows for the effective separation of MPI components, achieving direct ToF (dToF)- like high accuracy using a low-cost iToF-based sensor. Furthermore, the reconstruction network and the coded shutter are jointly optimized in an end-to-end framework to maximize depth estimation accuracy while explicitly considering shutter control constraints.

発表文献
IEEE Sensors Journal
吉田道隆
吉田道隆
博士後期課程学生
長原一
長原一
教授

コンピューテーショナルフォトグラフィ、コンピュータビジョンを専門とし実世界センシングや情報処理技術、画像認識技術の研究を行う。さらに、画像センシングにとどまらず様々なセンサに拡張したコンピュテーショナルセンシング手法の開発や高次元で冗長な実世界ビッグデータから意味のある情報を計測するスパースセンシングへの転換を目指す。