goto contents

Martin Nyolt

Efficient human situation recognition using Sequential Monte Carlo in discrete state spaces

Universität Rostock, 2019

Abstract: This dissertation analyses these challenges and provides solutions for SMC methods. The large, categorical and causal state-space is the largest factor for the inefficiency of current SMC methods. The marginal filter is analysed in detail for its advantages in categorical states over the particle filter. An optimal pruning strategy for the marginal filter is derived that limits the number of samples.

doctoral thesis   free access