Efficient Global Optimization of Non-Differentiable, Symmetric Objectives for Multi Camera Placement

We propose a novel iterative method for optimally placing and orienting multiple cameras in 3D scene. Sample applications include improving the accuracy of reconstruction, maximizing covered area surveillance, or coverage multi-viewpoint pedestrian tracking. Our algorithm is based on block-coordinate ascent combined with surrogate function an exclusion technique. This allows to flexibly handle difficult objective functions that are often expensive quantized non-differentiable. The solver globally convergent easily parallelizable. show how accelerate optimization by exploiting special properties function, such as symmetry. Additionally, we discuss trade-off between non-optimal stationary points cost reduction when optimizing viewpoints consecutively.