Experimental sensitivity investigation of mean-shift variants to bandwidth parameters for 3D buildings reconstruction

The goal of this work is to investigate on the one hand the sensitivity of three variants of the joint mean shift algorithm to the choice of the bandwidth parameters and on the other hand their upsampling capabilities in the area of 3D buildings reconstruction. In particular, we examined the quality of upsampled 3D building models obtained by fusing high resolution image data with low resolution elevation data acquired from airborne scanners from densely built urban areas. Using the root mean square error (RMSE) criterion computed between the actual upsampled result and a ground truth, prototype, 3D model, we provide both qualitative and quantitative comparative results. Discussion following the experimental section gives some insight to the observation that the spatially restricted joint mean shift seems to outperform the other two variants.