XARXES project at RNE

The project XARXES, a collaboration of CVC with the Centre for Demographic Studies, appeared last Saturday at Spain’s national radio broadcaster: RNE (Radio Nacional de España), in the program dedicated to the dissemination of Mathematics Raíz de 5 and presented by mathematician and communicator Santi García Cremades. In the clip, Dr. Oriol Ramos and Pau Riba explained the project and its implications for the search of our past. Listen to the whole podcast here (in Spanish): http://mvod.lvlt.rtve.es/resources/TE_SRA5IZZ/mp3/7/3/1524852288537.mp3 

 

Related article: XARXES: Connecting The Lives Of Our Ancestors

Vintra: AI for effective video surveillance

To solve a home invasion crime, the Police Services of Dublin (California) would had to review 70 hours of video obtained from a security camera from a neighbour’s home. This amount of footage would have meant more than three days of intense review for the team. Instead of that, they tested Vintra’s solution, which removed idle frames from static footage and showed them the crucial scenes. As a result, these 70 hours became 24 helping detectives to identify the suspect’s vehicle faster.

Although 24 hours meant an important reduction, Vintra’s founders assured that nowadays, this could be reduced to 6 hours due to the improvements the technology has experienced.

Vintra at its beginning

Vintra was born as a company at the end of 2016. It was founded by Brent Boekestein and CVC researchers Dr. Angel Sappa and Dr. Ariel Amato. They all presented certain concerns about the new deep learning techniques in the field of video analytics and therefore, starting with this first idea, they decided to create a prototype and a previous product which was successfully validated by final users.

Vintra was not the first business that these two CVC researchers set up. They first co-founded a CVC spin-off called Crowdmobile S.L together with Dr. Felipe Lumbreras; in that company they developed Knowxel, which is a kind of social network where users work in different tasks. These tasks need to be solved by a large team in a short period of time and are usually related to the need of companies and organisms to collect, analyse and process large amounts of data. With Knowxel, not only were companies able to complete their tasks swiftly but users were receiving money for the work done.

Back to Vintra, it is an AI software that helps security professionals get their job done. Vintra works with footage from all kinds of cameras and is capable to identify a wide array of key attributes within a video, including descriptive, definitive, scene and object attributes.

As this technology reduces the amount of attention required by people in charge of monitoring, it allows an improvement in terms of efficiency getting investigators to focus on other higher value tasks and money and resources spent in better analysis and resolution of cases. In fact, in the first example mentioned, the analysis of multiple day footage, allowed Dublin’s police to discover that the suspect had been casing the house with the same car three days prior to the robbery, and were also able to connect his vehicle with multiple cases in the surrounding areas. This would have been impossible without the help of this intelligent device.

The inaccuracy of manual reviews  

Cameras have a great presence in multiple cities around the globe. They are used to monitor streets and traffic, to generate a sense of security and are key to crime solving.  Cameras provide a vast amount of footage that plays a main role in the fight against delinquency.

Nonetheless, the fact of having an extensive amount of hours of video available, can actually be a downfall when it comes to solving a crime hurriedly. In fact, a single investigator spends between 200 and 300 hours per year reviewing video footage looking for a specific person, object or event.

Humans are not well prepared to do this task efficiently because of our limited ability for long-term singular focus. After a few minutes of video review, fatigue and video blindness cause not only distractions but cancels the viewer’s visual perception.

As Vintra’s founders explain, the device is not designed with the intention of replacing the investigator’s job, but it seeks to be a complementary tool that allows them to work more efficiently and focus attention on higher value tasks.

The future of Vintra

Nowadays, Vintra is formed by a team of 13 people. However, as Dr. Angel Sappa explained, their intention is to increase this number in the near future: “The future of Vintra is to continue growing and break into new markets. In fact, in medium term, we are evaluating the incorporation of new professionals in the field of machine learning”.

Moreover, Artificial Intelligence gets smarter and is able to recognize smaller differences the more it is presented with examples. Because of this, the device will keep improving, so that anything, from types of clothing or bags, to behaviours and emotions will be quickly and accurately searchable. Further still, it will be possible to digitally search real-time video and create alerts for people, objects, vehicles and behaviour.

For more information: https://vintra.io/

Building communities and innovation networks with robotics at schools – Steam Conference 2018

Dr. Fernando Vilariño and Alexandra Canet presented a workshop at last week’s Steam Conference in Cosmocaixa, Barcelona. They presented the Library Living Lab to High School science and technology teachers and dedicated a session on how to build a community with local actors and thus actively work on societal challenges within their towns or neighbourhoods, all of this with the use of robotics. The public was highly commited with their local communities and eager to explore different ways to effectively engage with stakeholders and thus trigger change at a local scale.

CVC Researchers at this year’s DAS 2018 Conference in Vienna

CVC presented a total of three papers and one demo at this year’s Document Analysis Conference that took take place in Vienna the 24th and 25th of April.  Find the papers here:

D. Karatzas, L. Gómez, M. Rossinyol, (2018): The Robust Reading Competition Annotation and Evaluation Platform

M. Carbonell, M. Villegas, A. Fornes, J. LLados (2018): Joint Recognition of Handwritten Text and Named Entities with a Neural End-to-end Model

L. Gómez, M. Rossinyol, D. Karatzas (2018): Cutting Sayre’s Knot: Reading Scene Text without Segmentation. Application to Utility Meters

VIDEOS

Our Associate Director, Dr. Dimosthenis Karatzas, explains his paper in this video:

Our PhD student Manuel Carbonell explains his paper in this video:

Dr. Lluis Gómez explains his paper in the following clip:

ORAIN: from small talk to machine talk

Nowadays, humans are in continuous contact with all kinds of machines. Washing machines, ticket sellers or snack vending machines, for example, are produced to make our tasks effortless and thus solve ordinary problems. However, the majority of these devices are absolutely passive, human – machine interaction being unidirectional and limited. So, what if these machines could communicate just like humans? What about if you could complain to them that your coin just got stuck – again!- or that your coffee isn’t quite of your taste that day?

That is precisely what Orain is trying to achieve. Born in 2005, Orain is a CVC spin-off and its aim is to introduce intelligence in our everyday life machines so that we can talk to them. Xavier Sans, one of its founders, explains that Orain was created almost by chance: “it emerged due to the demand to change a sector which was outdated and in urgent need of evolution. Only two years ago, there was not even the idea of paying on a machine with your smartphone. That’s when we saw a market niche in which we could grow”.

Their beginnings were hand in hand with another company, taking care of all the possible risks. Today, the team is configured by 9 employees (and rapidly growing) with different professional profiles such as developers and computer vision experts along with people in marketing and finance.

How does it work?

Orain allows communication between a vending machine and a smartphone through an application. Users can find information about its products such as ingredients, nutritional values or allergen components. Moreover, they can pay for the product instantly solely with their phones, be it by credit card or PayPal, and maintain a control of their expenses, thus providing an easier, quicker and enhanced user experience.

The benefits of this technology are also for clients. Whereas users obtain a highly positive shopping experience, vending operators receive detailed information about the state of their machines and relevant analytics: habits and consumption patterns, helping them to design effective marketing strategies and increase user loyalty. Furthermore, with this information, they can set up special campaigns or individual promotions in order to boost sales.

The innovative future of Orain

The concept Orain has designed goes far beyond payment methods and consumption analytics. It is thought to become a recurrent system platform with which to pay daily services such as car cleaning, washing machines, tickets, or periodical supermarket purchases. For now, Orain’s technology is operating in almost 300 vending machines and in 50 Office Coffee Services; a number that is rapidly increasing. Its technology will also be implemented in other sectors­ within the near future: “This is Orain’s strength. We want to develop a transversal application that allows us to gain users. We want one sole application to be used everywhere”, as stated by Xavier Sans.

Nevertheless, there’s still much to do. The startup is currently working in improving its services, with a special focus on Natural Language Processing: users speaking to the machine and obtaining an answer: “When you have a human interlocutor you can complain and do certain things that, in any other way, are impossible with machines. For this, we are trying to get close to human interaction”, Mr. Sans explains.

Orain within a human-machine interaction scenario

One of the challenges that Orain has to face is society’s reaction when faced with a human-machine interaction scenario. Sometimes, innovation within the AI field causes controversial opinions and, in some cases, generates fear and insecurity. In this way, the public acceptance of common machines equipped with intelligence can be, at the least, intimidating. “We are trying to make this transition the plainest possible. We have started with simple implementations, such as vending and OCS, but the intention is adding new intelligence step by step. We can’t be disruptive at the beginning because we can lose clients”, as pointed out by Mr. Sans.   

Another way to solve it is normalizing this kind of applications. As Xavier Sans predicts: “The solution is being the standard, which means being implemented massively. If we reach that point, it will be easier for people to comprehend and engage with our technology”.

For more information: orain.io