High Quality Photometric Reconstruction using a Depth Camera

In this work, we develop a depth-guided photometric 3D reconstruction method that works solely with a depth camera like the Kinect. Our method uses the in-built IR camera in such devices to obtain high quality normal estimates using photometric stereo and then from a depth-guided 3-D reconstruction of the object obtainable in each view, uses an effective multiview technique to reconstruct the full 3-D model of the object.


Full Pipeline

Figure 1. Full Reconstruction Pipeline


Figure 2 below shows the raw mesh obtained from a single depth map of Ganesh figurine from the Kinect. Figure 3 shows a single view reconstruction using 5 IR images and a depth map from a single view. Figure 4 shows the 3-view reconstruction using our method.

Single View Kinect Raw

Figure 2. Raw Kinect Mesh

Single View Photometric Reconstruction

Figure 3. Photometric Single View Reconstruction

Multiple View Photometric Reconstruction

Figure 4. Photometric Multiple View Reconstruction

Figure 5 below shows the full reconstruction using the raw Kinect depth maps of Buddha figurine. Figure 6 shows views of full reconstruction of the same using our method.

Reconstruction with Raw Kinect Depth maps

Figure 5. Reconstruction with Raw Kinect Depth maps

Our Reconstruction

Figure 6. Our Reconstruction

Figure 7 below shows the fine details recovered using our method from novel views.

Novel Rendered Views

Figure 7. Novel Rendered Views

Publications

  1. High Quality Photometric Reconstruction from a Depth Camera (Sk. Mohammadul Haque, Avishek Chatterjee, Venu Madhav Govindu), In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014. [bibtex] [pdf] [poster]