Enhancing the Street Geometry Extraction by Fusion of Thermal/Visible Images for Visual Servoing

Streets’ specifications such as width, orientation, and shape are useful and robust for localization applications. Extracting such specifications in case of bad weather or low-light might be not possible sometimes. This work suggests fusing visible images with thermal images to enhance the process of extracting the streets’ specifications. The monochrome threshold-based image fusion algorithm is adopted. The suggested method is tested on data collected in urban from a real flight for DJI drone equipped with both thermal and daylight cameras. After image fusion and specifications extraction, visual servoing on the street center line takes place. The results proved the fused image efficiency in low-light conditions.