Fast multi-camera reconstruction and surveillance with human tracking and optimized camera configurations

Abstract

Human-robot cooperation and coexistence requires robust surveillance. Our system creates a visual hull as volume approximation of humans within the work cell from images of multiple static color cameras. We present a fast, exact, and conservative reconstruction from high-resolution input data and apply temporal filtering after the reconstruction to address detection failure of the cameras. Furthermore, we avoid incorrect placement of cameras by automatically optimizing the camera positions. All processing steps handle occluding obstacles.