Real-time spotting of human activities in industrial environments
Real-time spotting of human activities in industrial environments
复制标题
实时发现工业环境中的人类活动
DOI：
--
复制
发表时间：
2008
期刊：
影响因子：
--
通讯作者：
T. Stiefmeier
中科院分区：
文献类型：
--
作者：
T. Stiefmeier
如今，几乎每个工作场所都有计算技术，以支持人们有效地完成工作。台式计算机、笔记本电脑和个人数字助理是这种技术的广泛例子。然而，工业环境中的生产和装配工人不能依赖这种传统设备。他们需要用双手来完成所谓的主要任务，例如组装复杂系统或维护精密机械。但与此同时，他们可以从计算机系统中受益，该系统为工人提供有用的信息。需要新的人机交互方式来解决这个问题。一种方法是上下文感知计算机系统，其检测用户的当前需求并预测他的未来需求以主动地为用户采取行动。研究人员已经证明，新兴的可穿戴计算技术有潜力在这一领域提供解决方案。然而，时至今日，可穿戴计算在工业应用中的部署仍然有限。本论文的目标是为可穿戴计算和情境感知系统在工业环境中的部署做出贡献。活动识别在设计用于工业环境的上下文感知计算机系统中起着重要作用。工人的运动携带了许多关于在这种环境中执行的活动的信息。在这项工作中，我们使用可穿戴惯性传感器不显眼地集成到夹克，以获取工人的运动活动识别。可穿戴计算平台由于功率、尺寸和重量的约束而具有有限的计算资源。因此，我们开发了一个连续的活动定位方法，具有较低的计算复杂度。该方法利用字符串匹配技术，需要一个离散的运动表示，我们从几个肢体轨迹。我们模型的活动类与模板字符串的活动定位操作过程中的连续运动字符串进行比较。融合阶段结合分布式信息，同时处理的身体轨迹和过滤阶段验证发现的活动发生。我们的方法的关键属性是，它结合了连续数据的分割和类，
Nowadays, computing technology is found in almost every workplace to support people in doing their job efficiently. Desktop computers, notebooks and personal digital assistants are widespread examples for this technology. However, production and assembly workers in industrial environments cannot rely on such traditional devices. They need to use their hands to carry out what is called a primary task, for example the assembly of a complex system or the maintenance of sophisticated machinery. Yet at the same time they could benefit from access to a computer system that provides the workers with useful information. New ways of human-computer interaction are required to solve this problem. One approach are context-aware computer systems that detect the user’s current needs and anticipate his future demands to proactively act for the user. Researchers have demonstrated that the emerging wearable computing technology has the potential to offer solutions in this field. However, to this day, the deployment of wearable computing in industrial applications is limited. The objective of this thesis is to contribute to the deployment of wearable computing and context-aware systems in industrial environments. Activity recognition plays a prominent role in designing context-aware computer systems for industrial environments. The motion of a worker carries much information about the performed activities in such environments. In this work, we use wearable inertial sensors unobtrusively integrated into a jacket to acquire the worker’s motion for activity recognition. Wearable computing platforms have limited computational resources due to constraints of power, size and weight. Therefore, we develop a continuous activity spotting method with low computational complexity. The method makes use of string matching techniques requiring a discrete representation of motion which we derive from several body limb trajectories. We model activity classes with template strings that are compared with the continuous motion string during the activity spotting operation. A fusion stage combines distributed information from simultaneously processed body trajectories and a filtering stage validates spotted activity occurrences. The key property of our method is that it combines the segmentation of continuous data and the clas-