Joint Multi-Task Multi-Attribute Learning 3D Behavior Recognition Based On Multi-Modal Data

3D behavior recognition has broad application prospects in the fields of image retrieval,intelligent monitoring,human-computer interaction,etc.It is a research hotspot in the field of artificial intelligence in recent years,Through the unremitting efforts of scholars at home and abroad,it has made great progress,However,due to the non-ideal imaging conditions and the complex and varied motion content,there are still many problems to be solved in 3D behavior recognition,such as the lack of uniform and effective feature description of action sequences,the semantic gap between low-level image features and high-level action categories,and lack of Reliable semantic feature time domain analysis models,etc.In view of the above problems,it is proposed to carry out research in two aspects.1)the multi-mode data characteristics of RGBD based on action sequences,and study the multi-space-time expression of 3D actions obtained through multi-convolutional neural network.2)research using graph model to describe the semantic and geometric relations between action visual attributes,constrain multi-task joint semantic attribute optimization learning,eliminate semantic gap.Finally we can realize complex behavior recognition of 3D action sequences.