A DSM adjustment method using a reference-based correction approach, Readar Case Study

Abstract

Readar is a company that provides aerial imagery and maintains large repositories of Digital Surface Models (DSM) stored as raster images. A DSM is an elevation map which includes height measurements of elevated objects (such as buildings or trees) . DSMs have several applications, such as identifying obstacles for aviation, vegetation management around power lines, and urban planning Methods to acquire DSMs are LiDAR and stereo imaging among others, the trade-off between cost and accuracy between these methods is well known \cite{dsm_concep_zhou}. DSMs LiDAR being both accurate and costly, are collected every three years over the Netherlands Stereo DSMs are captured every year but may be affected by vertical errors due to temporal changes or mismatches in sensors In this work, we developed a method that reduces the stereo DMS discrepancies by leveraging a reference LiDAR dataset. The approach relies on subsetting pixel values by triangle thresholding, and background segmentation to create a correction surface. The effectiveness was assessed by comparing original and adjusted DSMs through visual inspection, and volume computation which decreased about 10%. Finally, we experimentally evaluated the computational efficiency of the method, which requires a time comparable to the baseline step of writing the results. In addition, it does not require external parameters or intensive model training, which suitable for even larger datasets while providing explainable results.