We propose a method for 3D video at ~8 frames per second with 32 × 32 pixels resolution to be applied in hyperspectral camera single-pixel and RADAR for operating in outdoor applications with severe weather conditions or GPS-denied environments. Dynamic environments present different factors that limit performance in the quality image and video of conventional cameras based on RGB sensors as atmospheric effects, background light noise saturation, low-visibility, or obstacle on the scene, for which have been proposed different solutions from the integration of sensors hyper-spectral and fusion technologies as RADAR, and LIDAR. A low-cost solution for image sensors in low-light scattering media that offer more integration capacity with other sensors is a vision system based on the principle of single-pixel imaging (SPI). For that, we developed an SPI camera with active NIR illumination at wavelength 1550 nm in combination with millimeter RADAR in band 80 GHz for the generation of 3D video. For generation 3D video, we will implement an architecture GPU and define a reconstruction parallelized robust 3D algorithm based on the Shape-for-Shading (SFS) method, deep learning, and depth mapping with depth with RADAR to reach a near-continuous real-time 3D frame. As an experimental setup, we build a scenario to simulate different conditions environment as low-illumination, scattering, and irregularity surfaces, where we can evaluate the performance of our system vision in comparison with RGB camera using figure-of-merit as PSNR, RMSE, and VIF. Furthermore, to determine the system's capacity for future sensor vision by UAVs.
