Xerox Shows Progress In Machine Vision And Understanding: Gallery
Green technologies and machine vision were on display at the Xerox Research Centre in Grenoble
Xerox Research has gifted us with plenty of technology, from the humble mouse to the laser printer. It’s now working on ways to make machines see, and help people handle massive data sets - and it presented some of the fruits of this labour at an open day in Grenoble, France.
Systems designed to streamline city transport and to monitor medical patients without intrusive probes were just some of the projects seen by TechWeekEurope in a French chateau which continues the Xerox Research tradition. It all began at the Palo Alto Research Center (PARC), which led the world in the 1970s and 1908s, when it invented, among other things, the Ethernet and windowing software.
Research in a changing world
Times have changed. Nowadays, most people know of Xerox as a printer firm, but many are unaware that in 2009 it bought services firm ACS for $6.4 billion, and now employs more than 80,000 people in call centres and other service-related activity. The firm’s services revenue has grown steeply and now it makes up half the comapny’s turnover.
Meanwhile the research has continued, with teams of academics and experts plugging away at intractable tech problems in centres around the world. As well as PARC, Xerox now has research centres in Europe (Grenoble), Canada, New York and India.
The arrival of ACS has given the researchers a whole new set of problems – and they are clearly enjoying them. Xerox Research Centre Europe played host to a group of journalists this month, and the scientists, led by research officer Dr Monica Beltrametti (pictured) were clearly having a ball.
Overall, the experiments had a very strong “green” tint: Xerox is using tech to reduce printing, to cut the energy wasted in office systems, to help people park, and to make public transport more efficient.
Meanwhile, Luca Marchesotti wants to help computers sort and choose images based on their aesthetic characteristics. Xerox has developed an image “signature” and his system can distinguish “professional” quality images from “amateur” ones, then help find the image which fits with the colour scheme and style of a publication or web page.
Analytics and natural language
Research fellow Christopher Dance uses Big Data and analytics to expand the number problems that ACS’s transport division can handle. He and his colleagues showed us a system which is revolutionising parking in Los Angeles. Instead of having a fixed and uniform price for parking, the system – which covers 7000 parking spaces in an 800 block area – allows prices to change up and down, at intervals of around three months, in response to data on how the spaces have been used.
“Market rates means that parking spaces go to the people who value them most,” he explained. The prices are set to ensure around 85 percent of spaces are occupied, and therefore people who need to park only need to drive less than a block to find a space. For those who are not satisfied with the price, a cheaper zone will be a block or two away. The prices are set based on detailed reporting of actual parking data and – so far – it seems to be working.
In another room, Caroline Privault (pictured) reaches across a giant touch screen, to sift trough thousands of documents relating to a legal case. The data set is labelled “Enron”, and there’s some irony given that Xerox had its own accounting scandal twelve years ago. But Privault – a machine learning and linguistics expert – is looking at future cases.
The technology looks cool – but the hardware turns out to be straightforward, nothing more than a large TV mounted horizontally and fitted with a digitising bezel that can distinguish up to 30 points of touch, as a team of legal experts swipes their way through a stack of documents.
Legal teams are often handed thousands of emails, and need to quickly assess their relevance to a case. With the Enron documents, that could be done by sifting out those with specific key words, or sent by particular people, on particular dates.
Images for health and transport
Raja Bala’s research might make hospital life less painful for many patients, including premature babies. Instead of attaching intrusive probes, he wants to have patients monitored by video cameras which can measure pulse, temperature and lung volume, simply by analysing the differences between one video frame and the next.
Searching through images is also the main concern of Edgar Bernal. He wants to make it easier to preserve, and then find and extract the relevant data from CCTV cameras by compressing it according to the information videos contain. Half an hour of video from a traffic camera might contain 54000 frames. It is stored in compressed form, which normally uses frames at a fixed interval as reference images. To find images where vehicles pass means reconstructing the original video from the compressed form, and whizzing through it till you find the cars.
Instead, Bernal suggests finding the cars first, via image processing on-the-fly, and using those frames as reference images, so searching simply consists of looking at the 100 or so images where cars are visible.
Analytics is used again when Xerox starts to play on its home turf. Scientists, including ethnologist Jacki O’Neil, analyse work patterns by watching people in the office, while printer management software monitors the energy use of their printers. The result is a two-pronged attack. The printers are placed in the best position for people to use them – and they are turned off at times when the monitoring software reveals they are not used.
Does Xerox really want to reduce the number of pages printed? Absolutely, Tom Blodgett, the COO for Xerox Services in Europe told TechWeekEurope. The number of pages printed is going down, and Xerox would have no more future than Kodak if it tied itself to consumables. Like all printer firms, Xerox says it is shifting towards “document services” – and it is possible that it says this with more authority than some, given the strength of its research.
Many of these findings will have an impact on jobs – and journalists experienced a collective frisson, when told of a natural language parsing system which could extract basic facts from a set of input. Specifically, the project aims to take a lot of unstructured data, and answer the questions: Who, What, Where, When and Why.
That concerned us, as it is the definition of journalism. Xerox Research is some way off from achieving its goal, but it looks to be setting its sight on our jobs.
Are you a tech patent expert? Try our quiz!