【摘要】 提出一种基于局部自相似描述符和图集鲁棒性多任务学习的多视图交互行为识别方法。首先,提出了一种组合交互特征表示方法,它既编码了兴趣点局部运动的空间分布,又编码了上下文信息。此外,采用时间金字塔词袋模型描述自相似矩阵的局部特征,减小了观测角度变化对识别的影响,保留了时间信息。为了探索不同交互行为与不同视图之间的潜在相关性,采用图集鲁棒性多任务学习函数学习对应的交互行为识别模型。实验结果表明,该方法在公共数据库CASIA上与其他先进方法相比在交互行为识别中具有良好的识别效果。更多还原

【Abstract】 This paper proposed a novel multi-view interactive behavior recognition method based on local self-similarity descriptors and graph robust multi-task learning. First, we proposed the composite interactive feature representation which encodes both the spatial distribution of local motion of interest points and their contexts. Furthermore, local self-similarity descriptors represented by temporal-pyramid BOW was applied to decrease the influence of observation angle change on recognition and retain the temporal information. For the purpose of exploring latent correlation between different interactive behaviors and different views, graph robust multi-task learning penalized by image set constraint and robust model decomposition was used to learn the corresponding interactive behavior recognition model. Experiment results showed the effectiveness of the proposed method in comparison with other state-of-the-art methods on the public databases CASIA for interactive behavior recognition.更多还原