Sustainable work aims at improving working conditions to allow workers to effectively extend their working life. In this context, occupational safety and well-being are major concerns, especially in labor-intensive fields, such as construction-related work. Internet of Things and wearable sensors provide for unobtrusive technology that could enhance safety using human activity recognition techniques, and has the potential of improving work conditions and health. However, the research community lacks commonly used standard datasets that provide for realistic and variating activities from multiple users. In this article, our contributions are threefold. First, we present VTT-ConIoT, a new publicly available dataset for the evaluation of HAR from inertial sensors in professional construction settings. The dataset, which contains data from 13 users and 16 different activities, is collected from three different wearable sensor locations.Second, we provide a benchmark baseline for human activity recognition that shows a classification accuracy of up to 89% for a six class setup and up to 78% for a sixteen class more granular one. Finally, we show an analysis of the representativity and usefulness of the dataset by comparing it with data collected in a pilot study made in a real construction environment with real workers.

Existing work in human activity detection classifies physical activities using a single fixed-length subset of a sensor signal. However, temporally consecutive subsets of a sensor signal are not utilized. This is not optimal for classifying physical activities (composite activities) that are composed of a temporal series of simpler activities (atomic activities). A sport consists of physical activities combined in a fashion unique to that sport. The constituent physical activities and the sport are not fundamentally different. We propose a computational graph architecture for human activity detection based on the readings of a triaxial accelerometer. The resulting model learns 1) a representation of the atomic activities of a sport and 2) to classify physical activities as compositions of the atomic activities. The proposed model, alongside with a set of baseline models, was tested for a simultaneous classification of eight physical activities (walking, nordic walking, running, soccer, rowing, bicycling, exercise bicycling and lying down). The proposed model obtained an overall mean accuracy of 77.91% (population) and 95.28% (personalized). The corresponding accuracies of the best baseline model were 73.52% and 90.03%. However, without combining consecutive atomic activities, the corresponding accuracies of the proposed model were 71.52% and 91.22%. The results show that our proposed model is accurate, outperforms the baseline models and learns to combine simple activities into complex activities. Composite activities can be classified as combinations of atomic activities. Our proposed architecture is a basis for accurate models in human activity detection.