We propose PanopticFusion, a novel online volumetric semantic mapping system at the level of stuff and things. In contrast to previous semantic mapping systems, PanopticFusion is able to densely predict class labels of a background region (stuff) and individually segment arbitrary foreground objects (things). In addition, our system has the capability to reconstruct a large-scale scene and extract a labeled mesh thanks to its use of a spatially hashed volumetric map representation. Our system first predicts pixel-wise panoptic labels (class labels for stuff regions and instance IDs for thing regions) for incoming RGB frames by fusing 2D semantic and instance segmentation outputs. The predicted panoptic labels are integrated into the volumetric map together with depth measurements while keeping the consistency of the instance IDs, which could vary frame to frame, by referring to the 3D map at that moment. In addition, we construct a fully connected conditional random field (CRF) model with respect to panoptic labels for map regularization. For online CRF inference, we propose a novel unary potential approximation and a map division strategy. We evaluated the performance of our system on the ScanNet (v2) dataset. PanopticFusion outperformed or compared with state-of-the-art offline 3D DNN methods in both semantic and instance segmentation benchmarks. Also, we demonstrate a promising augmented reality application using a 3D panoptic map generated by the proposed system.

This paper presents a study of a method for estimating the position and orientation of a photo-shoot in indoor environments for augmented reality applications. Our proposed localization method is based on registered frame data of virtualized reality models, which are photos with known photo-shoot positions and orientations, and depth data. Because registered frame data are secondary product of modeling process, additional works are not necessary to create registered frame data especially for the localization. In the method, a photo taken by a mobile camera is compared to registered frame data for the localization. Since registered frame data are linked with photo-shoot position, orientation, and depth data, 3D coordinates of each pixel on the photo of registered frame data is available. We conducted experiments with employing five techniques of the estimation for comparative evaluations.

Recently, generation of novel views from images acquired by multiple cameras has been investigated. It can be applied to telepresence effectively. Most conventional methods need some assumptions about the scene such as a static scene and limited positions of objects. In this paper, we propose a new method for generating novel view images of a dynamic scene with a wide view, which does not depend on the scene. The images acquired from omni-directional cameras are first divided into static regions and dynamic regions. The novel view images are then generated by applying a morphing technique to static regions and by computing visual hulls for dynamic regions in real-time. In experiments, we show that a prototype system can generate novel view images in real-time from live video streams.

Recently, generation of novel views from images acquired by multiple cameras has been investigated in the fields of virtual and mixed reality. Most conventional methods need some assumptions about the scene such as a static scene and limited positions of objects. We propose a new method for generating novel view images of a dynamic scene with a wide view, which does not depend on the scene. The images acquired from omni-directional cameras are first divided into static regions and dynamic regions. The novel view images are then generated by applying a morphing technique to static regions and by computing visual hulls for dynamic regions in realtime. In experiments, we show that a prototype system can generate novel view images in real-time from live video streams.