Skills: Python, bash, PyQt5 & Designer, Docker, Git, CI/CD, CUDA, OpenDroneMap, OpenSfM, OpenMVS, WSL, MeshLab, GDAL, GIS.
Description: In this project, I developed an algorithm to project bounding boxes in image pixel coordinates (u,v) into their corresponding global coordinates (X,Y) in the desired projection system (eg. EPSG:32611, or equivalently WGS 84 / UTM zone 11N) . This open-source tool I built takes as inputs (1) a nadir drone image with drone location and pose metadata available in its EXIF header (2) a digital surface model DSM, (3) the image coordinates to be projected as a CSV and the desired projection system. Based on principles of 3D computer vision, my algorithm outputs an ESRI shapefile containing the coordinates of the projected polygons.
Batch functionality and reverse projection direction (GIS to Image, i.e., shapefile of geo-referenced polygons to image local coordinates) have also been implemented! :)
In the image below, the red box indicates the ground-truth bounding box for the red tree (denoted by the orange dot in the center), while the green box is the projected polygon from the drone image-space coordinate system for the same tree (as visible in the image).