Efficient generation of occlusion-aware multispectral and thermographic point clouds

The reconstruction of 3D point clouds from image datasets is a time-consuming task that has been frequently solved by performing photogrammetric techniques on every data source. This work presents an approach to efficiently build large and dense co-acquired images. In our case study, the sensors co-acquire visible as well thermal multispectral imagery. Hence, RGB are reconstructed with traditional methods, whereas rest sources lower resolution less identifiable features projected into first one, i.e., most complete dense. To this end, mapping process accelerated using Graphics Processing Unit (GPU) multi-threading in CPU (Central Unit). accurate colour aggregation points guaranteed taking account occlusion foreground surfaces. Accordingly, solution shown reconstruct much more than notable commercial software (286% average), e.g., Pix4Dmapper Agisoft Metashape, time (−70% average respect best alternative).