Computational Imaging for Scene Understanding

2Transient Convolutional Imaging

Felix HEIDE

Princeton Computational Imaging Lab, Princeton University, USA

2.1. Introduction

Inspired by biological sonar, optical time-of-flight sensors measure the delay between light emitted into a scene and received by a co-located sensor. It was the invention of the laser in 1960, as well as accurate timing electronics, that allowed for a commercially successful application of this principle in Lidar systems. While delivering high accuracy and range, a major drawback is that distance is only measured to a single point, and hence, scanning is required. Recently, correlation image sensors (Lange 2000) have been revolutionizing the 3D depth imaging market by temporally convolving a scene with amplitude-modulated flood illumination, which no longer requires scanning. Correlation image sensors convolve an incoming temporal signal with a reference signal. From this convolutional measurement, the time-of-flight can be estimated with high accuracy. As such time-of-flight systems do not require a baseline or lack of textured objects, they outperform structured light or stereo matching techniques by a significant margin for scenes with large range. Correlation image sensor technology has been developed since the early 2000s, and adoption has been rapidly accelerated by sensor fabrication in the CMOS process. CMOS technology has been driven by the microprocessor industry and is therefore inexpensively available for mass-market production.

Adding ...