We present a novel hierarchical model for human activity recognition. In contrast with approaches that successively recognize actions and activities, our approach jointly models actions and activities in a unified framework, and their labels are simultaneously predicted. The model is embedded with a latent layer that is able to capture a richer class of contextual information in both state-state and observation-state pairs. Although loops are present in the model, the model has an overall linear-chain structure, where the exact inference is tractable. Therefore, the model is very efficient in both inference and learning. The parameters of the graphical model are learned with a structured support vector machine. A data-driven approach is used to initialize the latent variables; therefore, no manual labeling for the latent states is required. The experimental results from using two benchmark datasets show that our model outperforms the state-of-the-art approach, and our model is computationally more efficient.
DOCUMENT
Vertical urbanisation is perceived as necessary to accommodate a growing population but is associated with severe risks for human well-being. It requires a profound understanding of how archi-tectural designs can ensure visually readable and liveable environments before it has been built. How-ever, current digital representation techniques fail to address the diverse interests of non-experts. Emerging biometric technologies may deliver the missing user information to involve (future) inhabit-ants at different stages of the planning process. The study aims to gain insight into how non-experts (visually) experience 3D city visualizations of designed urban areas. In two laboratory studies, univer-sity students were randomly assigned to view a set of the same level of detail images from one of two planned urban area developments in the Netherlands. Using eye-tracking technology, the visual behav-iour metrics of fixation count and duration and general eye-movement patterns were recorded for each image, followed by a short survey. The results show how visual behaviour and perception are remark-ably similar across different detail levels, implying that 3D visualizations of planned urban develop-ments can be examined by non-experts much earlier in the design process than previously thought.
DOCUMENT
The use of machine learning in embedded systems is an interesting topic, especially with the growth in popularity of the Internet of Things (IoT). The capacity of a system, such as a robot, to self-localize, is a fundamental skill for its navigation and decision-making processes. This work focuses on the feasibility of using machine learning in a Raspberry Pi 4 Model B, solving the localization problem using images and fiducial markers (ArUco markers) in the context of the RobotAtFactory 4.0 competition. The approaches were validated using a realistically simulated scenario. Three algorithms were tested, and all were shown to be a good solution for a limited amount of data. Results also show that when the amount of data grows, only Multi-Layer Perception (MLP) is feasible for the embedded application due to the required training time and the resulting size of the model.
DOCUMENT
