Online Multi-Task Clustering for Human Motion Segmentation

摘要: Human motion segmentation in time space becomes attractive recently due to its wide range of potential applications on action recognition, event detection, and scene understanding tasks. However, most existing state-of-the-arts address this problem upon an offline single-agent scenario, while there are a lot urgent requirements segment videos captured from multiple agents for real-time application (e.g., surveillance system). In paper, we propose Online Multi-task Clustering (OMTC) model online multi-agent where each agent corresponds one task. Specifically, linear autoencoder framework is designed project sequences into common motion-aware across collaborating tasks, the decoder obtains representation task via temporal preserved regularizer. To tackle distribution shifts between pair task-specific projections further proposed align By way, significant knowledge can be shared among data structures also well preserved. For optimization, efficient effective optimization mechanism derived solve large-scale formulation applications. Experiment results Keck, MAD our collected human datasets demonstrate robustness, high-accuracy efficiency OMTC model.