基于人体动作识别与理解的人机交互是实现人机智能装配的关键，从数学角度进行的人体动作识别与理解的研究泛化能力不足，而目前的仿生视觉研究对先验知识作用机制的分析与模拟尚有欠缺。本项目以人类视皮层视觉感知机制与大脑皮层理解机制为切入点，通过分析视觉信息前馈连接和反馈连接机制，实现人体装配动作识别与理解的仿生建模，解决自底向上的视觉特征信息和自顶向下的指导作用的融合机制以及多维度视觉特征信息和记忆知识的匹配机制两个难题。具体内容包括：通过一种视觉显著度映射关系来模拟自底向上的视觉特征信息和自顶向下的指导作用的融合机制，构建基于生物视觉激励的装配动作识别仿生模型；基于视觉特征信息仿生模型，通过一种特征向量关联模型来模拟多维度视觉特征信息和记忆知识的匹配机制，构建生物理解机制启发的装配动作理解仿生模型。

The human-computer interaction based on human action recognition and understanding is the key to human-computer intelligent assembly. The generalization ability of the study on human action recognition and understanding from the mathematical view is insufficient, while the analysis and simulation of mechanism of prior knowledge in the current research of bionic vision are still insufficient. The visual perception mechanism of human visual cortex and the understanding mechanism of human cerebral cortex are taken as the starting point in this project. By analyzing the feedforward connection and feedback connection mechanism of visual information, a bionic model of human assembly action recognition and understanding is established to solve the fusion mechanism of bottom-up visual feature information and top-down guidance, as well as the matching mechanism of multi-dimensional visual feature information and memory knowledge. The specific contents include: Simulating the fusion mechanism of bottom-up visual feature information and top-down guidance through a visual saliency mapping relationship to build a biological vision stimulation based bionic model of assembly action recognition; simulating the matching mechanism of multi-dimensional visual feature information and memory knowledge through a feature vector correlation model to build a biological understanding mechanism inspired bionic model of assembly action understanding.