Optimal Viewpoint Selection by Indoor Drone Using PSO and Gaussian Process With Photographic Composition Based on KL Divergence

This study processes an autonomous indoor drone photographer that searches for and selects a heuristic optimal viewpoint to obtain a well-composed photograph of a group of subjects. The subjects on the drone’s camera screen are represented by a Gaussian mixture model. When there are four or more subjects, they are represented by a Gaussian mixture model with clustering by variational Bayes. The Kullback–Leibler divergence is evaluated between the Gaussian mixture model and a user-defined reference composition, and it is defined as the composition evaluation value. The reference composition is pre-set by the user based on the basic composition rules, such as the three-section method. The drone searches for a viewpoint in a 3D space to optimize the composition evaluation value using particle swarm optimization (PSO). A Gaussian process is used to facilitate the PSO search. This enables the drone to significantly reduce the search time and successfully capture a photograph with a well-balanced composition.