Conceptually, each sensor image helps one to assign to each point in the 3D grid a likelihood that this grid point is on the surface. To this end, the object points get projected into the sensor image, and a correlation figure-of-merit gets computed at that image location. Since one has multiple overlapping images sensor views, one will get multiple estimates at each grid point. The surface is then defined by an optimization process using both the computed likelihoods at each grid point from image matches, as well as constraints about the surface’s smoothness, thus information about the object. The mathematical formulation of the point cloud generation is found in Zach [55], Irschara et al. [54] and Pock et al. [56].