42 P Er Va Si V E Blockincomputing Wearable Activity Tracking in Car Manufacturing

W earable computing systems should provide unobtrusive access to the IT infrastructure, so they should be physically unobtrusive, leave users’ hands free, and allow freedom of motion. In an industrial scenario, they must also minimize workers’ cognitive load and avoid distracting them from their primary tasks (see the sidebar “Related Work in Wearable Computing in Industrial Environments”). An appropriate user interface design can partially accomplish this. A better solution, however, is a proactive context-aware system that uses unobtrusive sensors to track each step of the performed task and presents the worker with the information needed at any given moment. Little research has focused on complex activity recognition for industrial applications. In this field, system reliability and robustness are essential. We’re working to develop and test such real-life industrial activity tracking systems as part of the European Union’s 4.5year-long wearIT@work project.1 To this end, we conducted two case studies in cooperation with the European car manufacturer Skoda. Our first study involved context-aware worker training at a dedicated learning island, which consisted of an instrumented car body. The second considered wearable assistance for quality control in the production line, where workers use on-body sensors only. After observing and videorecording workers in the Skoda factory, we made the first data recordings onsite. Because such recordings disrupt the production process, causing large costs, we recreated the environments in our lab using original tools, cars, and procedures. We brought an entire car body, identical to the Skoda learning island, into our premises. Also, Skoda provided us with a complete new car on which to reenact the quality-control steps from the videos and observations. Based on experiments performed in these setups over the last two years, we present an overview of the challenges and approaches associated with using complex multimodal sensor systems for activity tracking in production environments.