Request access For more information
Computer Vision, Machine Learning and Pattern Recognition

Object recognition

Responsable: Àgata Lapedriza, David Masip 

Object recognition in images is still one of the most important research topics in computer vision. Given an image or a video, the goal of object recognition is to recognize and localize all the objects.

In the last recent years, this topic has experienced an impressive gain in performance with the use of Deep Neural Networks [1] and big datasets such as ImageNet [2]. Despite of the research efforts, object recognition is an unsolved problem. For the methods that perform in real time (such as Deformable Part Models [3]), the detection accuracy is low, while the methods that show higher performance can not run in real time. Actually, even the best current algorithms for object recognition are still far away from human performance. In this research line we focus on improving current systems, both in terms of accuracy and speed.

For more information please see or contact

[1] A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classification with deep convolutional neural networks. In In Advances in Neural Information Processing Systems, 2012.

[2] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierar- chical image database. In Proc. CVPR, 2009.

[3] P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan. Object Detection with Discriminatively Trained Part Based Models. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 32, No. 9, Sep. 2010.

Scene recognition and understanding

Responsable: Àgata Lapedriza

Understanding complex visual scenes is one of the hallmark tasks of computer vision. Given a picture or a video, the goal of scene understanding is to build a representation of the content of a picture (e.g., what are the objects inside the picture, how are they related, if there are people in the picture what actions are they performing, what is the place depicted in the picture, etc).

With the appearance of large scale databases like ImageNet [1] and Places [2], and the recent success of machine learning techniques such as Deep Neural Networks [3], scene understanding has experienced a large amount of progress, making possible to build vision systems capable of addressing some of the mentioned tasks.

In this research line, in collaboration with the computer vision group at the Massachusetts Institute of Technology, our goal is to improve existing algorithms for scene understanding and to define new problems that become reachable now, thanks to the recent advances in neural networks and machine learning.

For more information please see or contact

[1] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierar- chical image database. In Proc. CVPR, 2009.

[2] B. Zhou, A. Lapedriza, J. Xiao, A. Torralba, and A. Oliva. "Learning Deep Features for Scene Recognition using Places Database." Advances in Neural Information Processing Systems 27 (NIPS), 2014.

[3] A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classification with deep convolutional neural networks. In In Advances in Neural Information Processing Systems, 2012.

Recognition of facial expressions

Responsable: David Masip 
Facial expressions are very a important source of information for the development of new technologies. Humans communicate our emotions through our faces, and Psychologists have studied emotions in faces since the early works of Charles Darwin [1]. One of the most successful emotion models is the Facial Action Coding System (FACs [2]), where a particular set of action units (facial muscle movements) act as the building blocks of 6 basic emotions (happiness, surprise, fear, anger, disgust, sadness).

The automatic understanding of this universal language (very similar in almost all cultures) is one of the most important research areas in computer vision, with applications in many fields, such as design of intelligent user interfaces, human-computer interaction, diagnosis of disorders or even the reactive publicity field. In this research line we propose to apply and design state-of-the-art supervised algorithms to detect and classify emotions and Action Units.

Nevertheless, there exist far more emotions in addition to this basic set. We can predict with better than chance accuracy: the results of a negotiation, the preferences of the users in binary decisions [3], the deception perception, etc. In this research line we collaborate with the Social Perception Lab from Princeton University ( to apply automated algorithms to real data from Pyschology Labs.

For more information please see or contact

[1] Darwin, Charles (1872), The expression of the emotions in man and animals, London: John Murray.

[2] P. Ekman and W. Friesen. Facial Action Coding System: A Technique for the Measurement of Facial Movement. Consulting Psychologists Press, Palo Alto, 1978.

[3] Masip D, North MS, Todorov A, Osherson DN (2014) Automated Prediction of Preferences Using Facial Expressions. PLoS ONE 9(2): e87434. doi:10.1371/journal.pone.0087434

Human Pose Recovery and Behavior Analysis

Responsable: Xavier Baró 
Human Action/Gesture recognition is a challenging area of research that deals with the problem of recognizing people in images, detecting and describing body parts, inferring their spatial configuration, and performing action/gesture recognition from still images or image sequences, also including multi-modal data. Because of huge space of human configurations, body pose recovery is a difficult problem that involves dealing with several distortions: illumination changes, partial occlusions, changes in the point of view, rigid and elastic deformations, or high inter and intra-class variability, just to mention a few. Even with the high difficulty of the problem, modern Computer Vision techniques and new tendencies deserve further attention, and promising results are expected in the next years.

Moreover, several subareas have been recently defined, such as Affective Computing, Social Signal Processing, Human Behavior Analysis, or Social Robotics. The effort involved in this area of research will be compensated by its potential applications: TV production, home entertainment (multimedia content analysis), education purposes, sociology research, surveillance and security, improved quality live by means of monitoring or automatic artificial assistance, etc.

For more information please see or contact

Biologically-inspired computer vision

Responsable: David Masip 
The complex network that forms the human brain allows us to recognize thousands of objects, actions and scenes in a few milliseconds with almost no effort. The vision problem can be considered solved in the biological brain, but we are far from achieving satisfying solutions in computational systems. In this topic we propose to develop computational algorithms inspired in the human brain, or more specifically, in the ventral visual processing stream. Previous research has shown that a specifically tuned set of bank of filters can mimic the V1 IT cortex (inferior temporal cortex) [1]. Nevertheless, the "untangle" process that obtains meaningful information from "pixel" images in the retina remains unknown. We will propose deep learning architectures [2] to obtain robust computational algorithms applied to visual tasks in the machine.

The algorithms developed will be applied to real computer vision problems applied to Neuroscience, in the Princeton Neuroscience institute, and range from detection and tracking of rodents in low resolution videos, image segmentation and limb detection, and motion estimation of whiskers using high speed cameras.

For more information please see or contact

[1] Serre, Thomas, Lior Wolf, Stanley Bileschi, Maximilian Riesenhuber, and Tomaso Poggio. "Robust object recognition with cortex-like mechanisms."Pattern Analysis and Machine Intelligence, IEEE Transactions on 29, no. 3 (2007): 411-426.

[2] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." In Advances in neural information processing systems, pp. 1097-1105. 2012.

Distributed, Parallel and Collaborative Systems

Parallel and Distributed Scientific Applications: Performance and Efficiency

Responsable: Josep Jorba
Currently, the growth in parallel and distributed programming paradigms and environments, available for performing concurrent computations, has different bottlenecks for providing efficient applications.

We need to know the platforms, their performance, the underlying hardware and networking technologies, and we must be able to produce optimized software that statically or dynamically may take advantage of the computational resources available.

In this research line, we study different approaches to do better scientific applications, and to make tools (via automatic performance analysis), which can understand the application model and the underlying programming paradigm, and try to tune their performance to a dynamically changing computational environment, in which the resources (and their characteristics) can be homogeneous or heterogeneous depending on the hardware platform. In particular we focused our research on shared memory and message-passing paradigms, and in many/multi-core environments including multi-core CPUs, GPUs (graphic cards computing) and cluster/grid/cloud/super computing platforms.



Reliability and availability issues in distributed computer systems

Responsable: Joan Manuel Marquès
Increasingly, services are being deployed over large-scale computational and storage infrastructures. To meet ever-increasing computational demands and reduce both hardware and system administration costs, these infrastructures have begun to include Internet resources distributed over enterprise and residential broadband networks. As these infrastructures increase in scale to hundreds of thousands to millions of resources, issues of resource availability and service reliability inevitably emerge.

Contributive distributed systems based on non-dedicated resources provide attractive benefits but face at least one significant challenge: nodes can enter and leave the collective on a whim, because each machine may be separately owned and managed. At the same time, resource availability is critical for the reliability and responsiveness (low response latency) of services. Groups of available resources are often required to execute tightly coupled distributed and parallel algorithms of services. Moreover, load spikes observed with Internet services require guarantees that a collection of resources is available.

The goal in this research line is to determine and evaluate predictive methods that ensure the availability of a collection of resources. We strive to achieve collective availability from non-dedicated resources distributed around Internet, arguably the most unreliable type of resource worldwide. Our predictive methods could work in coordination with a virtualization layer for masking resource unavailability.

More details:

Community cloud systems based on non-dedicated resources

Responsable: Joan Manuel Marquès
Many online communities exist nowadays: social networks, open source development, Wikipedia, Wikileaks etc. These communities have thousands or even millions of participants and are commonly hosted in resources belonging to entities not directly related with participants in the community. Considering that most of the participants' computers are underutilised - i.e. have more resources than are actually used in daily activity - we are studying how to take advantage of these surplus resources to be able to host communities using only the resources provided by their members. We name these kinds of communities contributory communities. More precisely, we are working on a) availability prediction of non-dedicated resources; b) privacy guarantees, especially anonymity; and c) efficient deployment and storage.

Therefore, we are looking for researchers interested in large-scale distributed systems applied to contributory communities in fields such as availability guarantees, privacy, efficient deployment of services or reliable storage.

More details:

Simulation environments for networking and distributed systems

Responsable: Joan Manuel Marquès
Large-scale testbeds like PlanetLab open new opportunities to test and simulate network protocols, distributed systems or applications. An open issue in these kinds of systems is how to create, deploy, execute and monitor experiments in a coordinated way, in order to take advantage of the real deployment of systems and protocols. A great deal of research needs to be undertaken to understand the kind of systems that can benefit from these environments and the systems that are better simulated in classical simulation environments. Adapting existing simulation environments to these new systems is not a trivial process and a great deal of work has to be done in order to characterize the distributed environment and its coordination needs. We are especially interested in building simulation environments for distributed systems or networking in large-scale testbeds both for research and teaching.

More details:

Time Synchronized Channel Hopping (TSCH)

Responsable: Xavi Vilajosana
The revolution of operational technologies in industries is being lead by a new generation of low power wireless standards rooted at a technique known as Time Synchronized Channel Hopping (TSCH). TSCH was designed to allow IEEE802.15.4 devices to support a wide range of applications including, but not limited to, industrial ones . At its core is a medium access technique which uses time synchronization to achieve ultra low-power operation and channel hopping to enable high reliability. Synchronization accuracy impacts power consumption, and can vary from micro-seconds to milli-seconds depending on the solution.

IEEE802.15.4e is the latest generation of ultra-lower power and reliable networking solutions for LLNs. [RFC5673] discusses industrial applications, and highlights the harsh operating conditions as well as the stringent reliability, availability, and security requirements for an LLN to operate in an industrial environment. In these environments, vast deployment environments with large (metallic) equipment cause multi-path fading and interference to thwart any attempt of a single-channel solution to be reliable; the channel agility of TSCH is the key to its ultra high reliability. Commercial networking solutions are available today in which nodes consume 10's of micro-amps on average with end-to-end packet delivery ratios over 99.999%.

IEEE802.15.4e TSCH focuses on the MAC layer only. This clean layering allows for TSCH to fit under an IPv6 enabled protocol stack for Low Power Lossy Networkss, running 6LoWPAN [RFC6282], IPv6 Routing Protocol for Low power and Lossy Networks (RPL) [RFC6550] and the Constrained Application Protocol (CoAP) [RFC7252]. What is missing is a Logical Link Control (LLC) layer between the IP abstraction of a link and the TSCH MAC, which is in charge of scheduling a timeslot for a given packet coming down the stack from the upper layer.

While [IEEE802154e] defines the mechanisms for a TSCH node to communicate, it does not define the policies to build and maintain the communication schedule, match that schedule to the multi-hop paths maintained by RPL, adapt the resources allocated between neighbor nodes to the data traffic flows, enforce a differentiated treatment for data generated at the application layer and signaling messages needed by 6LoWPAN and RPL to discover neighbors, react to topology changes, self-configure IP addresses, or manage keying material.

The research line aims to study mechanism to attain efficient resource utilization, allocation and management in low power wireless sensor networks in the context of the IEEE802.15.4e and the initiative. It is focused on the study of scheduling techniques to access to the wireless medium taking into account energy consumption and energy availability amongst others. The line also aims to push forward the management capabilities of such constrained networks by developing the concept of differentiated services, label switching and resource reservation in LLNs. The topic is currently being studied at the IETF 6TiSCH working group where the candidates will contribute their results.

Simulation and Optimization

Optimization and Simulation of Industrial and Engineering Systems

Responsable: Angel A. Juan 
Internet Computing & Systems Optimization (ICSO) is an IN3 official programme supported by the DPCS research group. One of the main research topics in ICSO is the development of new hybrid algorithms and methods which combine applied optimization (e.g. heuristics and metaheuristics), discrete-event simulation, and data analysis to support decision-making processes in realistic environments. In particular, we are interested in real-life applications of these algorithms to logistics, transportation and production systems. Thus, the PhD theses will be related to any of the following topics: Rich (real-life) Vehicle Routing Problems, real-life Scheduling Problems, Green Logistics, Intelligent Transport Systems, Horizontal Collaboration in Logistics, etc. These topics represent important challenges for industrial sectors in any developed country, which explains their relevance in the context of current international research.

Webs: (HAROSA KC), (ICSO Programme),

Information and Communication Systems and Services

Outsourcing of ICT services for companies and public administrations

Responsable: Josep Maria Marco-Simó 
After undertaking the in-depth case study of the Framework Agreements for the Standardisation of ICT Services for the Government of Catalonia, which constituted the first PhD thesis of the group, ICSS intends to continue working in the wider research area of outsourcing of ICT services by both private companies and public administrations, including universities.

Procurement and implementation services for integrated systems

Responsable: Joan Antoni Pastor 
We continue working in the analysis and development of procurement and implementation methodologies for integrated information systems such as those based upon customer relationship management (CRM) systems, supply chain management (SCM) systems, enterprise application integration (EAI) projects and information systems integration in general, as well as enterprise resource planning (ERP) systems. We also include within our interests along these lines the analysis and development of models and methods for decisional information systems such as those based on business intelligence technology.

Novel approaches for ICT services management and governance

Responsable: Joan Antoni Pastor 
Our work and interest focuses on the analysis and development of models and methods for managing ICT services, informed by the principles of IT governance and agile operations. To this effect, one chief aim of the group is to come to grips with, and offer solutions to, the main professional problems found in the domain of strategic management and planning of information systems and ICT services.

Innovative curriculum engineering for ICT programmes at any level

Responsable: Maria Jesus Marco-Galindo 
Although probably more fit for other UOC PhD programmes such as "Education and ICT", we are interested in researching novel ICT curriculum experiences and approaches at any level, including university and vocational studies, which we aim to do in collaboration with other disciplines such as pedagogy. We already have an advanced PhD thesis on analysis of the case of UOC's innovation on communicative written stills for ICT students, and the design and development of a new transversal system-service in this area.

Novel formal approaches and methods for researching ICT and IS

Responsable: Joan Antoni Pastor 
Usually orthogonal to and integrated with other research lines, we are explicitly interested in experimentation and publication with suitable research methods for investigating problems in our field (case studies, action research, grounded theory, design science research, PLS).