Figure 2From: Novel Kernel-Based Recognizers of Human ActionsFor supervised action recognition, for each sequence we compute a single feature vector (left) representing histogram-based data, as described in Section 6. Such feature vectors are then mapped to an RKHS by using a novel kernel which is both characteristic and appropriates for said representation (right). An SVD classifier is learnt from labeled training sequences in order to classify different actions.Back to article page