Context-Adaptive Heterogeneous Models for Human Activity Recognition

在广泛的现实世界应用的推动下，最近人们做出了重大努力来探索基于数据的人类活动识别模型，该模型利用AP适配器收集的信息。这些数据一般是从日常环境中收集的，其中包含许多阻碍人类活动识别的环境特有因素。另一方面，数据分布在计算能力和电池续航时间不同的采集设备上。这两个挑战促使研究人员设计上下文自适应的异质模型，该模型在理论上和经验上对多场景下的人类活动识别都是有效的，并且可以以最小的努力推广到新的上下文。可持续能源需求对模型轻量化的可扩展性带来了另一个限制。所提出的模型将使用Widar3.0等开放数据集和测量数据进行验证。

Driven by a wide range of real-world applications, significant efforts have recently been made to explore the data-based human activity recognition model that utilizes the information collected by the AP adaptor. The data is generally collected from the daily environment, which contains a lot of environment-specific factors hampering the recognition of human activities. On the other hand, the data is distributed at collection devices with different computing capabilities and battery life. These two challenges motivate the research to design context-adaptive heterogeneous models that are theoretically and empirically effective for human activity recognition in multiple scenes and can generalize to a new context with minimal effort. The sustainable energy requirement introduces another constraint on the scalability of models to be lightweight. The proposed models will be verified using open datasets such as widar3.0 and the measured data.