Automatic 3D reconstruction of Indoor Environments from Consumer-graded Cameras
With the increase demand of 3D models in urban environment, automatic indoor modeling has attracted more attention for human trajectory identification, facility management and indoor navigation. To reconstruct indoor models quickly and reliably, this project will propose a new indoor reconstruction method to automatically generate indoor semantic building models through building ground maps and images from consumer-graded cameras. The image of building ground maps will be over-segmented and later semantically classified as ceilings, walls and other indoor elements. The 3D models of indoor elements will be later reconstructed from extracted geometric information and semantic labeling. Finally, multiple images from consumer-graded cameras will be registered with the generated model. The advantage of this proposed method is that it fully combines semantic information from building ground maps and texture from multiple images. Therefore, it will automatically reconstruct indoor models with lower cost.