Abstract

Abstract

Abstract
Despite the high performance of current presentation attack detection (PAD) methods, the robustness to unseen attacks is still an under addressed challenge. This work approaches the problem by enforcing the learning of the bona fide presentations while making the model less dependent on the presentation attack instrument species (PAIS). The proposed model comprises an encoder, mapping from input features to latent representations, and two classifiers operating on these underlying representations: (i) the task-classifier, for predicting the class labels (as bona fide or attack); and (ii) the species-classifier, for predicting the PAIS. In the learning stage, the encoder is trained to help the task-classifier while trying to fool the species-classifier. Plus, an additional training objective enforcing the similarity of the latent distributions of different species is added leading to a 'PAI-species'-independent model. The experimental results demonstrated that the proposed regularisation strategies equipped the neural network with increased PAD robustness. The adversarial model obtained better loss and accuracy as well as improved error rates in the detection of attack and bona fide presentations. © 2019 Gesellschaft fuer Informatik.

Abstract
Developing autonomous assistants to help with domestic tasks is a vital topic in robotics research. Among these tasks, garment folding is one of them that is still far from being achieved mainly due to the large number of possible configurations that a crumpled piece of clothing may exhibit. Research has been done on either estimating the pose of the garment as a whole or detecting the landmarks for grasping separately. However, such works constrain the capability of the robots to perceive the states of the garment by limiting the representations for one single task. In this paper, we propose a novel end-to-end deep learning model named GarmNet that is able to simultaneously localize the garment and detect landmarks for grasping. The localization of the garment represents the global information for recognising the category of the garment, whereas the detection of landmarks can facilitate subsequent grasping actions. We train and evaluate our proposed GarmNet model using the CloPeMa Garment dataset that contains 3,330 images of different garment types in different poses. The experiments show that the inclusion of landmark detection (GarmNet-B) can largely improve the garment localization, with an error rate of 24.7% lower. Solutions as ours are important for robotics applications, as these offer scalable to many classes, memory and processing efficient solutions.

Abstract
The automotive industry is currently focusing on automation in their vehicles, and perceiving the surroundings of an automobile requires the ability to detect and identify objects, events and persons, not only from the outside of the vehicle but also from the inside of the cabin. This constitutes relevant information for defining intelligent responses to events happening on both environments. This work presents a new method for in-vehicle monitoring of passengers, specifically the task of real-time face detection in thermal images, by applying transfer learning with YOLOv3. Using this kind of imagery for this purpose brings some advantages, such as the possibility of detecting faces during the day and in the dark without being affected by illumination conditions, and also because it's a completely passive sensing solution. Due to the lack of suitable datasets for this type of application, a database of in-vehicle images was created, containing images from 38 subjects performing different head poses and at varying ambient temperatures. The tests in our database show an AP50 of 99.7% and an AP of 78.5%.

Abstract
The present demographic trends point to an increase in aged population and chronic diseases which symptoms can be alleviated through rehabilitation. The applicability of passive 3D reconstruction for motion tracking in a rehabilitation context was explored using a stereo camera. The camera was used to acquire depth and color information from which the 3D position of predefined joints was recovered based on: kinematic relationships, anthropometrically feasible lengths and temporal consistency. Finally, a set of quantitative measures were extracted to evaluate the performed rehabilitation exercises. Validation study using data provided by a marker based as ground-truth revealed that our proposal achieved errors within the range of state-of-the-art active markerless systems and visual evaluations done by physical therapists. The obtained results are promising and demonstrate that the developed methodology allows the analysis of human motion for a rehabilitation purpose.