Bingyao Huang — Page 2

arXiv2019

End-to-end Projector Photometric Compensation

Projector photometric compensation aims to modify a projector input image such that it can compensate for disturbance from the appearance of projection surface. In this paper, for the first time, we formulate the compensation problem as an end-to-end learning problem and propose a convolutional neural network, named CompenNet, to implicitly learn the complex compensation function. CompenNet consists of a UNet-like backbone network and an autoencoder subnet. Such architecture encourages rich multi-level interactions between the camera-captured projection surface image and the input image, and thus captures both photometric and environment information of the projection surface. In addition, the visual details and interaction information are carried to deeper layers along the multi-level skip convolution layers. The architecture is of particular importance for the projector compensation task, for which only a small training dataset is allowed in practice. Another contribution we make is a novel evaluation benchmark, which is independent of system setup and thus quantitatively verifiable. Such benchmark is not previously available, to our best knowledge, due to the fact that conventional evaluation requests the hardware system to actually project the final results. Our key idea, motivated from our end-to-end problem formulation, is to use a reasonable surrogate to avoid such projection process so as to be setup-independent. Our method is evaluated carefully on the benchmark, and the results show that our end-to-end learning solution outperforms state-of-the-arts both qualitatively and quantitatively by a significant margin.

cs.CV↗

arXiv2018

A Single-shot-per-pose Camera-Projector Calibration System For Imperfect Planar Targets

Existing camera-projector calibration methods typically warp feature points from a camera image to a projector image using estimated homographies, and often suffer from errors in camera parameters and noise due to imperfect planarity of the calibration target. In this paper we propose a simple yet robust solution that explicitly deals with these challenges. Following the structured light (SL) camera-project calibration framework, a carefully designed correspondence algorithm is built on top of the De Bruijn patterns. Such correspondence is then used for initial camera-projector calibration. Then, to gain more robustness against noises, especially those from an imperfect planar calibration board, a bundle adjustment algorithm is developed to jointly optimize the estimated camera and projector models. Aside from the robustness, our solution requires only one shot of SL pattern for each calibration board pose, which is much more convenient than multi-shot solutions in practice. Data validations are conducted on both synthetic and real datasets, and our method shows clear advantages over existing methods in all experiments.

cs.CV↗