In this paper we study the problem of estimating innercyclic time intervalswithin repetitive motion sequences of top-class swimmers in a swimming channel.Interval limits are given by temporal occurrences of key-poses, i.e.distinctive postures of the body. A key-pose is defined by means of only one ortwo specific features of the complete posture. It is often difficult to detectsuch subtle features directly. We therefore propose the following method: Giventhat we observe the swimmer from the side, we build a pictorial structure ofposelets to robustly identify random support poses within the regular motion ofa swimmer. We formulate a maximum likelihood model which predicts a key-posegiven the occurrences of multiple support poses within one stroke. The maximumlikelihood can be extended with prior knowledge about the temporal location ofa key-pose in order to improve the prediction recall. We experimentally showthat our models reliably and robustly detect key-poses with a high precisionand that their performance can be improved by extending the framework withadditional camera views.
展开▼