Automatic texturing of 3D building models using airborne infrared image sequences

The objective of this research work is automatic texturing of 3D building models using airborne thermal infrared (TIR) image sequences. TIR images of building hull are often used for the detection of damaged and weak spots in the insulation of the building hull. However, imaging and analysing each face manually is very time consuming. This process can be carried out more efficiently using a mobile mapping system (MMS) equipped with a TIR video camera and GPS/INS system. TIR images of building façades are captured by a terrestrial TIR camera mounted on a moving vehicle and the roofs by an airborne TIR camera mounted on a UAV or helicopter. The façades in the inner yards which are not visible to the terrestrial camera are captured by airborne camera, thanks to the oblique looking angle of the airborne camera. The assignment of the image sequences to the buildings is achieved by texture mapping on 3D building models.

In this research project, the major emphasis is put on texture mapping using airborne oblique view TIR images. Approximated exterior orientation (ExtOri) parameters of the camera are known thanks to the GPS/INS system. Hence the 3D building models are projected into the image. However, the image structures do not match the projected building models very well. To improve the fit between the 3D building models and the image structures, a matching procedure is applied and ExtOri parameters are recalculated using least squares adjustment method. After the best fit was found, the visibility tests are carried out and textures for each face of the 3D building models are extracted. This procedure is repeated for each frame. Accordingly, for many faces multiple textures are available. To solve this issue for each texture a quality value based on the resolution and the occlusion factor is calculated and for each face the best texture is selected.

As a future scope of work, advance the matching procedure will be advanced. Currently only one frame is used to find the best ExtOri parameters. In the future the advantage of image sequences will be taken and neighbouring frames will be used to improve the search for correspondences and stabilise the adjustment.

 

Bild1

 

 

Links:

  • PIA19

    Photogrammetric Image Analysis 18.-20. September 2019 an der TUM
  • MRSS19

    Munich Remote Sensing Symposium 18.-20. September 2019 an der TUM
  • DGPF 2019

    DGPF-Jahrestagung 20.-22. Februar 2019 in Wien