A5053231231- Context-Aware Activity Recognition Systems
- Occupational Health and Safety Research
- Anomaly Detection Techniques and Applications
- Non-Invasive Vital Sign Monitoring
- Human Mobility and Location-Based Analysis
- Heart Rate Variability and Autonomic Control
- Green IT and Sustainability
- Traffic and Road Safety
- Human Pose and Action Recognition
- Image Retrieval and Classification Techniques
- Video Surveillance and Tracking Methods
- Customer churn and segmentation
- Technology and Data Analysis
- Occupational Health and Safety in Workplaces
- Technology Adoption and User Behaviour
- Hemodynamic Monitoring and Therapy
- Laser-induced spectroscopy and plasma
- Image and Object Detection Techniques
- Target Tracking and Data Fusion in Sensor Networks
- Software System Performance and Reliability
- Software Testing and Debugging Techniques
- Recycling and Waste Management Techniques
- Indoor and Outdoor Localization Technologies
- IoT and Edge/Fog Computing
- Healthcare Technology and Patient Monitoring
VTT Technical Research Centre of Finland
2009-2024
Sustainable work aims at improving working conditions to allow workers effectively extend their 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 wearable sensors provide for unobtrusive technology that could enhance using human activity recognition techniques, has the potential health. However, research community lacks commonly used standard datasets realistic variating...
People use smartphones in daily activities for accessing and storing information various situations. In this paper, we present a work progress detecting automating some of these activities. To explore the possible patterns developed an experimental application to detect tasks used by analyzed it provide suggestions "routines". We conducted two-week user study with 10 users evaluate approach. During logged usage patterns, sent server where was analysed clustered. The participants could also...
This paper describes experiences and lessons learned in applying an approach of using a physical robot for testing overall smartphone device performance based on user profiles. The process consists capturing actions, abstracting them to usage profiles, transforming these into test models, generating cases from the models. goal is support touch screen devices applications realistic environment. To achieve this, tests are real-world generated profiles executed real robot, simulating actual...
Human Activity Recognition (HAR) from wearable sensor data identifies movements or activities in unconstrained environments. HAR is a challenging problem as it presents great variability across subjects. Obtaining large amounts of labelled not straightforward, since signals are easy to label upon simple human inspection. In our work, we propose the use neural networks for generation realistic and features using activity monocular videos. We show how these generated can be utilized, instead...
Many cloud and shadow detection methods have been proposed already, but improvements can be made on accuracy or automation. In this study, we propose a Fully Convolutional Network model for the of clouds shadows in optical satellite images. The was trained 165 Landsat images Finland, tested an independent set reached 95%, outperforming both quantitatively qualitatively selection other deep learning architectures.
Terrorism is an international security challenge. The early detection of threats (e.g., explosives or firearms) could provide a valuable contribution to the ability prevent, protect and respond terrorism. This paper presents system for management plurality sensors improve threat-detection capabilities without disrupting flow passengers. improves prevention soft targets (such as airports, undergrounds railway stations) with high number daily commuters. architecture consists three main...
Social networking service for mobile devices is presented and evaluated. The operating principle of the inspired by human-like cumulative gossiping. usage browser based implementation utilises standard ad hoc communication between smartphones. gossiping protocol built on level top existing method. evaluated with six social groups. Results suggest that there great demand this kind due to similarity human-type Also, a number propositions technical solutions further enhance end user experience...
Existing work in human activity detection classifies physical activities using a single fixed-length subset of sensor signal. However, temporally consecutive subsets signal are not utilized. This is optimal for classifying (composite activities) that composed temporal series simpler (atomic activities). A sport consists combined fashion unique to sport. The constituent and the fundamentally different. We propose computational graph architecture based on readings triaxial accelerometer....
Instance segmentation is a computer vision task where separate objects in an image are detected and segmented. State-of-the-art deep neural network models require large amounts of labeled data order to perform well this task. Making these annotations time-consuming. We propose for the first time, iterative learning annotation method that able detect, segment annotate instances datasets composed multiple similar objects. The approach requires minimal human intervention needs only...
Abstract Waste electronic and electric equipment (WEEE) are collected in high amounts the EU. However, order to enable safe effective recycling of their plastic fraction, harmful additives inside plastics need be identified. In this study, two spectroscopic methods, laser-induced breakdown spectroscopy (LIBS) Raman were employed for characterizing different brominated flame retardants (BFRs) from actual WEEE stream, also lab-made plastics. The results preliminary study indicate ability LIBS...
Human Activity Recognition (HAR) from wearable sensor data identifies movements or activities in unconstrained environments. HAR is a challenging problem as it presents great variability across subjects. Obtaining large amounts of labelled not straightforward, since signals are easy to label upon simple human inspection. In our work, we propose the use neural networks for generation realistic and features using activity monocular videos. We show how these generated can be utilized, instead...
In this article, we introduce a new approach for estimating the heart rate from noisy photoplethysmography (PPG) signals. We propose use of two-dimensional representations signals that are fed into residual deep neural network performs regression task. Our leverages transfer learning and pre-trained models to further reduce prediction error, resulting in state-of-the-art results challenging benchmark dataset.