Automatic annotation of human actions in video

This paper addresses the problem of automatic temporal annotation of realistic human actions in video using mini- mal manual supervision. To this end we consider two asso- ciated problems: (a) weakly-supervised learning of action models from readily available annotations, and (b) tempo- ral localization of human actions in test videos. To avoid the prohibitive cost of manual annotation for training, we use movie scripts as a means of weak supervision. Scripts, how- ever, provide only implicit, noisy, and imprecise information about the type and location of actions in video. We address this problem with a kernel-based discriminative clustering algorithm that locates actions in the weakly-labeled train- ing data. Using the obtained action samples, we train tem- poral action detectors and apply them to locate actions in the raw video data. Our experiments demonstrate that the proposed method for weakly-supervised learning of action models leads to significant improvement i...

• Type: Conference
• Year: 2009
• Where: ICCV
• Authors: Olivier Duchenne, Ivan Laptev, Josef Sivic, Francis Bach and Jean Ponce
• Keywords: Computer Vision, Human Actions, ICCV 2009, Video Annotation