| SEAMLESS CO-REGISTRATION OF MULTISPECTRAL CAMERA IMAGES BASED ON TRIFOCAL TENSOR ESTIMATION |
| Paper ID : 1108-SMPR |
| Authors: |
|
Camilo Cortes *1, Mozhdeh Shahbazi1, Ting Chan2 1Geomatics Engineering, Schulich School of Engineering, University of Calgary 2Guangdong Provincial Key Laboratory of Urbanization and Geo-simulation, School of Geography and Planning, Sun Yat-sen University |
| Abstract: |
| Since their development, miniature multi-camera multispectral systems aboard platforms with limiting weight and size restrictions, such as Unmanned Aerial Vehicles (UAVs), have become a popular choice for fine-scale mapping of vegetated areas. Commercially available photogrammetric software can process the image data collected by these sensors, producing ortho-rectified imagery and digital surface models. However, misalignment of several pixels between spectral bands has been observed to be a common issue when employing these solutions, which can translate to decimetre level ground distance errors, undermining the spectral and geometric analysis of the data. We propose a two-fold solution to seamlessly co-register the images of five cameras integrated in a multispectral camera system at each capture instance. This approach consists of 1) a robust self-calibration of the multispectral camera system to estimate the intrinsic calibration parameters and relative orientation parameters among all cameras; 2) a single image registration method based on trifocal constrains that maps all images into a common image space. This approach differs from previous methods, as it does not make any assumptions about the scene and does not use any best-fit projective or similarity transformations. Furthermore, it is not sensitive to large spectral variations between bands, as it does not attempt cross-spectral matching between bands with considerable visual differences. Results confirm the proposed co-registration method accurately fuses images from a multispectral sensor and is invariant to large depth variation in the captured scene. Automatic co-registration of images can be used to decrease the time complexity when processing large datasets. |
| Keywords: |
| multispectral, data fusion, camera calibration, mapping, trifocal tensor, co-registration |
| Status : Paper Accepted (Poster Presentation) |
