Focus on research: Dr Suzanne Little, Insight
Dr Suzanne Little is a funded professor with the media analytics group at the Insight Centre for Data Analytics based at DCU. In this interview she talks about the problems and opportunities presented by self-driving cars, the benefits of working with a stadium test bed and how to get more young people involved in STEM.
Cloud-LSVA covers a wide-ranging system for self-driving cars. How does this work?
Actually, Cloud-LSVA (Large Scale Video Analytics) is motivated by the need for semantic annotation of very high volumes of video – it just happens to be a big problem in the field of instrumented vehicles that generate huge quantities of recordings to improve driver assistance systems. An application of this could be to support development of self-driving cars but the main research in the project is about tools to enable the semi-automatic labelling of objects like cars, pedestrians, street signs, lanes etc in video. This involves applying state of the art methods for video analytics and running these in a cloud system. These videos are then used for developing new technologies and for testing that they are safe and comply with regulations. Without the automatic labelling and search facilities being developed by Cloud-LSVA, this is a very labour intensive and restricted task.
VI-DAS (Vision Inspired Driver Assistance Systems) is a separate project that I’m involved in and it looks more directly at technologies to improve driver safety by using sensors placed all around the car and driver. These can be used to create models of the situations around the car (other cars, lanes, signs, people, etc.) and of the interior of the car (driver alertness, distraction, direction of view) and combine these to understand and predict scenarios. A big challenge is understanding how to pass control of the vehicle back and forth with the driver rather than focussing on a fully self-driving (autonomous) car.
Some experts have said the technology for safe self-driving cars already exists and the only blockage to it becoming mainstream is fear. Would you agree with this or has the technology still a way to go before it can be considered reliable?
That has an element of truth in it, although ‘fear’ is a provocative term – lack of public understanding of technology is certainly an issue. I believe the technology is very advanced and has exciting current applications in specific circumstances such as closed lane transport systems (eg. bus corridors), factories, hospitals or campuses etc.
One non-technical issue preventing them from becoming widely used is certainly public acceptance. However, the unpredictability of putting fully self-driving systems into a public space with standard cars and pedestrians, who may not behave as expected, is also an ongoing challenge. Another important issue is the complexity of insurance and this is being explored by partners in Cloud-LSVA and VI-DAS. As technology developers we not only need to improve the performance of our algorithms but also to consider how they can be explained and measured.
You’ve worked as part of the team on the Croke Park IoT test bed. What projects are we seeing come out of there and what makes stadiums such useful labs for this kind of research?
There’s a range of projects that are happening in the Smart Stadium for Smarter Living (Smart Croke Park) test bed. The technical team from DCU, Intel, Microsoft and Croke Park have just filmed a video for the Microsoft Developers Network Channel 9 that discusses some of the projects and technical challenges we’ve faced.
We’ve got a number of sensors linked to Intel gateways that push data up to Azure. Last year we used two internal microphones that measure decibel levels to examine the sound of the ’16th player’ – loud the crowd got at certain points in the match (eg teams entering, goals scored). This was displayed on the big screen at halftime. We have ongoing work using video to measure and understand how crowds move around the stadium – near concession areas, etc. We’re using this to research new algorithms for crowd counting, activity recognition and exploiting processing power on the gateways and cloud platforms to improve performance and protect data by processing video close to the point of capture.
A camera is continuously collecting footage of the Croke Park pitch to understand and measure exposure to sunlight and various other sensors (wind speed, rainfall) are also being used. We are constantly coming up with new ideas – it’s a fantastic source of new data and inspiration.
Stadiums are great for IoT research as they are smaller in scale than a full Smart City deployment but much larger than a few sensors in a lab and so able to better test and prove technologies. It’s a more controlled environment than a city, with a defined boundary and known events, so you can set up and manage short term experiments more easily and predictably but it still has many of the same challenges as other environments. For example, how to move people around safely, securely and predictably during a range of events including sport, concerts, conferences and tourism.
There are opportunities for environmental monitoring and energy usage and to look at how the stadium operates within an urban area. There are also practical benefits including dealing with fewer stakeholders than other public areas. Stadium management have been very responsive and enthusiastic participants in the project and this helps with logistic challenges such as access to locations, electricity, data, people etc. Less practically there’s a certain romance, for want of a better word, that comes from working with Croke Park. The public has an investment in the space (‘a stadium of dreams’) and, we hope, that as the test bed develops that they can feel part of science and innovation too.
DCU is set to interduce a new undergraduate degree in Data Science. It seems there has been significant input from industry in developing the curriculum. Will this become the norm?
The new undergraduate degree in Data Science has had a lot of input from industry who are eager for new graduates to fulfil roles in their organisations. This has produced a degree with real practical benefits and that will have applied data problems to explore through specific modules in third and fourth year that can react to industry needs for certain skills. It’s a tricky balance in computing where technologies move incredibly quickly and, for example, the language of choice has changed from Cobol to Java to Python, so if a company wants a specific tool then computing courses can be too reactive and result in a course that doesn’t provide solid core skills that will future-proof graduates.
While it’s great to have input from industry to guide course content, it’s equally important to produce graduates with excellent foundational skills in programming, mathematics and data science so that they can adapt well when the latest and greatest new platform takes off. This technology hasn’t even been thought of yet.
DCU does a good job of working closely with and responding to companies so while I think that designing complete courses like the Data Science degree is unusual, we’ll likely continue to respond by adapting individual modules for industry needs.
You’ve also been involved in Girls Hack Ireland. Are you starting to see the benefits of these kind of outreach programmes in terms of the gender split in University admissions?
I think we’re a few years away from seeing the impact of programmes to increase the diversity of participation in STEM courses like computing. CoderDojo, for example, is an excellent initiative to get children programming and participants in that are still around three to five years from taking their leaving certificate. Girls Hack Ireland has been running for just three years so again we’re a little way off. I hope that the range of great programmes offered by places like Insight, DCU and SFI will start to normalise the idea of computing as a career and to give more people – girls and boys – experiences of technology to inspire them. I’m pleased to see that computing is now being developed as a Leaving Certificate subject.
I think we need to continue to invest in initiatives like the SFI Discover Programme and educate the general public about science. Parents have an important role to play in the choices their children make about university degrees so that’s why Girls Hack Ireland has tried to include them in our events as well.
Working with data there’s no getting away from issues of privacy and security. How do you see the General Data Protection Regulation changing your work?
This has always been an important consideration when working with data and most research groups I’ve worked with have tried to take steps to ensure that data (images, video, surveys) is handled appropriately.
It’s difficult to say exactly what the impact will be of the new regulations but we’ve already been working with the data protection officers at Insight and DCU to evaluate our projects and audit practises. My worry is that the fear of the new regulations and potential consequences will stifle the development of new public datasets in all areas by placing them without consideration in the ‘too hard’ basket.
For progress to be made in computer vision and machine learning (and many other areas) it’s necessary for shared datasets to be used to evaluate and compare ideas. That’s why initiatives like TRECVid, a large workshop that releases video datasets for teams to develop and evaluate new algorithms, have been so influential. Evaluation workshops will probably become more important as, by working together, research communities can ensure that GDPR is followed. Centres like Insight are also working to educate people on what data is and how it is used and this will help.