Stereo Matching for Terrain Mapping using Aerial Imagery

In recent years, the demand for accurate and detailed terrain mapping has escalated in the field of urban planning and environmental monitoring. Advanced techniques like Stereo matching using Semi-Global Matching (SGM) have been in use for years to generate depth maps. In this paper, the use of this algorithm is expanded to generate depth maps for various terrain regions using SGM on stereo pair of aerial imagery. Five variations are compared with each other namely linear stretching followed by Gaussian blur, dilation, bilateral filtering, guided filtering and morphological operations. after fine-tuning the SGM variables for different terrains (ground, city, river, mountains, etc.). Pre-processing techniques for color enhancement and smoothening on input aerial images have been tested to determine their usability, along with post-processing methods for noise reduction, occlusion handling and edge preservation in order to produce an accurate disparity map. The results have been evaluated using RMSEs values and visual analysis. The proposed methodology serves as an effective way to generate 3D models for complex terrain regions without the need of generating expensive ground truths providing up to 8% greater accuracy compared to existing SGM models for different terrain types.