HDR Video System Based on Multi-Stage Motion Estimation Using Dual Exposure Camera

In this paper, we provide a new HDR video system using a dual exposure camera. In order to prevent ghost artifacts caused by misalignment of motions, multi-stage convolutional neural network (CNN)-based motion estimation is proposed. In the proposed method, true motions and occlusion maps are estimated, and then an HDR video is synthesized with motion compensated images. It provides not only robustness against exposure levels but also flexibility to extend easily search ranges by stacking stages with same CNN coefficients. Experimental results indicate that the proposed method is effective to estimate true motions so that ghost artifacts are eliminated well.