Virtual Camera Composition with Particle Swarm Optimization

A Benchmark for Virtual Camera Control

“…For virtual camera optimisation in static environ-ments, e.g. for the generation of a single shot, there are two well established performance measures, which have been already adopted in a number of previous studies: best solution quality and convergence time [3,12]. The first measure expresses how good the algorithm is in finding the right camera configuration and, thus, generate a good quality shot.…”

“…A multitude of algorithms based, for instance, on genetic algorithms [11], particle swarm optimisation [12] or hill climbing [13] have been proposed and evaluated. However, while most of the articles composing the-state-of-the-art include an empirical evaluation of the algorithms, those evaluations have been carried out with different metrics and on different test problems.…”

“…The work by Burelli et al [12], for instance, includes an evaluation on three static camera control problems in an outdoor urban environment and lists convergence time and best solution quality as success criteria. Following a completely different approach, Lino et al [13] use an animated scene from the film 1984 on which the test multiple shots.…”

Automatic Camera Control: A Dynamic Multi-Objective Perspective

“…The different objective functions are combined linearly to produce a single objective function which can be optimised either in a static environment or in a dynamic one. A variety of algorithms have been employed in the two cases including, among others, Genetic Algorithms [15,17] and Particle Swarm Optimisation [6] for static scenes, and Hill Climbing [4] and Artificial Potential Fields [7] for real-time optimisation in dynamic scenes. The first two approaches are used to generate still images with specific composition characteristics, while the last two are designed to animate a camera in real-time interactive virtual environment -e.g.…”

Combining local and global optimisation for virtual camera control

“…Pure global optimisation approaches [23], [11], [8], [17] are capable of producing well composed shots with respect to designer requirements. However, their high computational cost makes them inappropriate for real-time interactive applications such as games.…”