【摘要】 为了把握国家“双减”政策导向，解决儿童居家学习时间延长却无人监督的矛盾，本文设计和实现了一种基于跨模态时空嵌入学习的儿童学习无人监督系统。该系统具有实时识别儿童动作、同步转播监控视频、设置学习任务并跟踪完成情况等功能。其中，动作识别功能基于跨模态时空嵌入学习模型实现，模型识别精度高且具有良好的迁移学习能力，在相关数据集上实现了高达95%的动作识别正确率。实践结果表明，该系统功能多样、动作识别准确、实时性强，能够推动儿童学习监督无人化、便捷化、高效化，有效节约父母的监督时间成本。更多还原

【Abstract】 In order to grasp the national "double reduction" policy orientation and solve the contradiction between extension of children’s home learning time and absence of parental supervision, this paper designs and develops an unmanned-supervision system for children’s learning based on multimodal spatio-temporal embedded learning. The system has the functions of real-time action recognition, synchronous broadcast of monitoring videos, setting learning tasks and tracking its completion. Among them, the action recognition function is based on the multimodal spatio-temporal embedded learning model, which has high recognition accuracy and eminent transfer learning ability. The action recognition accuracy is up to 95% on the relevant datasets. According to the experimental result, this system has diverse functions, accurate action recognition accuracy and strong real-time feedback, which helps to promote unmanned, convenience and efficiency for children learning supervision, and effectively saves parents’ supervision time cost.更多还原