We are pleased to present a video demonstration of the REUNITE system.
Much more information on how the system works is available in the report.
Even more information and the actual implementation can be found in the source.
REUNITE investigates how technology can help aid organisations reunite those separated by war and natural disaster.
In crisis, such as Boxing Day tsunami, communication channels are destroyed and families and friends are separated. Many flee to relief camps, where there may be more than a million people. In camps of this size, it’s almost impossible to find loved ones. So, aid organisations provide services to help people find those they’ve lost.
In a typical scenario, aid workers interview those who are looking for loved ones and asking for their personal details, the details of those they’ve lost and how they were separated. In a laborious process, this information is recorded onto paper. By manually searching these paper records, aid workers identify those who are looking for each other.
Aid workers try to reunite people as fast as possible. However, there may be few aid workers with the language skills required to conduct interviews and paper records are difficult to search and may contain errors.
Finding an alternative to this time-consuming process is the aim of REUNITE. We present a proof of concept that addresses the issues mentioned by replacing paper records with smartphones, increasing the workforce available using a concept called crowdsourcing and ensuring reliability with statically machine learning techniques.
We attended the Random Hacks of Kindness (RHOK) event at Digital Birmingham on the 4th/5th December 2010. RHOK is a 24 hour hackathon, where teams from all over the world help contribute their skills to solving some of the toughest humanitarian problems. Over the weekend we created an application called “Height Catcher”, more on which can be seen in the video below.
HeightCatcher from Reunite on Vimeo.
There was a great team of people who attended, a big thank you to James Cattell, Sarah Mount, Seb, Pete Smith, Monika Solanki, Nirmal Solanki, Danny Williams, and Sara Farmer for helping make for a fun-packed weekend.
From Left: Seb, Pete, Monika, Lloyd, Peter, James, Danny
The problem that Peter and I chose, “Height Catcher” , was based on a problem definition proposed by Cyrus Shahpar. It helps address the problem of identifying malnourishment of children within relief camps. Malnourishment is inferred from recording a child’s height and weight; in relief camps a large, heavy wooden board is used to measure a child’s height. A more rapid accurate assessment of height or length would be extremely helpful to assessing malnutrition in emergencies.
An example of a board used for measuring an infant's height
We worked through the night, the atmosphere was great and the other attendees were very friendly and accommodating. We ‘hacked’ our way through the night away, writing line after line code on a potent cocktail of tuna sandwiches and iced tea.
Nice, relaxing and therapeutic debugging, best served at 3am
We completed the experience with three hours sleep on the floor; it was not all in vain though, our perseverance paid off and by midday Sunday we had a working prototype which more or less made up for our stiff necks.
The app can be downloaded from here (best to do this from your android device, as it know what to do with .apk’s).
Photos of the event can be found here.
Source code for the android app can be found on github here.
You can also take a look at ShortCoords, one of the other projects created at RHOK Birmingham.
One of the major sticking points of our project is the concept of having a crowd extract information about a person from a rich media source.
Firstly let’s just clarify some terms.
So what we want is to present the crowd with a rich media source for them to use as a reference to help them fill in a form containing information on the person.
The problem is balancing the ‘richness’ of the media so to speak and the size of the media (in bytes). A large bandwidth (e.g. > 256 kbps upload) is something we can not just take for granted, especially if we are to create something which can be deployed in areas of low connectivity.
So when we came to considering recording the interview with a person seeking loved ones as video we had problem, we were going to be dealing with files of sizes too large for sensible use in the domain. We then investigated the compression options available on Android devices and were able to tweak it such that we got the size down such that the bit rate on the videos was around 75 kbps.
Yet when reviewed the benefits of video, the extra information gained from a video of low quality did not justify the overhead in size. What we’ve decided opt for now is recording the interview as audio and using a high quality (well higher quality than the images in the video) still image as the rich media source.
Over the next week we’re going to be performing some experiments to see how this works out, stay tuned.
Yesterday Pete and I did a presentation on the story behind our project so far and the technologies we have used at a PDC event in Manchester.
The event was a live streaming of the Microsoft Professional Developer Conference held in Seattle. It was hosted by the Manchester Entrepreneurs and Microsoft UK.
We presented a short overview of the history of our project so far and then talked about some of the technical solutions that we’ve been exploring for our project. We mentioned the benefits that the cloud platform offered a project such as ours, the key point being scalability, so when there is a big influx of traffic (i.e. after a disaster) we can scale our resources appropriately and in quieter times, shrink them down – keeping us efficient.
This fitted in well with the PDC event itself, which had a lot of exciting news for developers in regards to the cloud platform. There were announcement on the Azure platform’s new features, focussing on the ease of migrating existing systems to the cloud and offering integrations with existing systems with “public” and “private” clouds.
Last week Lloyd and I went to Birmingham to take part in the Beyond 2010 hackathon, it was facilitated by Digital Birmingham and ScraperWiki. We spent 24 hours straight working on ideas that would aid in disaster relief efforts. We had a great team that pushed on through the night, with Julian Todd, Aidan McGuire, Aine McGuire, Simon Whitehouse and Lloyd and I all working through to daybreak. We also had Sara Farmer from Open Crisis aid us in the design of the system, the results of which can be viewed in this wonderful diagram.
Here’s a video that better explains what we made over the 24 hours (and what we tidied up a little once we’d had some sleep)
Where’s Safe from Reunite on Vimeo.
The source code for the android application can be found here.
The code for the web service can be found here.
A key aim of this project is to create a system that allows tasks to be crowd sourced reliably.
At the moment, we’re in the early stages of planning a machine learning (ML) framework that will be able to oversee work done by a crowd. We call this framework Crowd Control.
One of the greatest challenges is obtaining crowd input in a form that can be easily used by a ML system. Over the past two days, I’ve been brain storming crowd input methods. I’ve made a few prototypes and I would like to share one of them with you.
A WikiString allows a crowd to work collaboratively on a piece free text, just like Wikipedia. However, a WikiString spends most of its time working out who did what and where each bit of text has come from. By analysing the changes made by the crowd, the hypothesis is that we’ll be able to infer when the text is complete or reliable. Here are some screen shot of it in action.
The screen shots show my testing tool. Each colour is a different author. The tool is only exposing a fraction of the information that it is collecting. I simply wanted to see how well the difference algorithm was preforming. It works quite well and I think I can make it work exceptional well with a few simple tweaks.
Reunite is a system that utilises crowdsourcing and machine learning techniques to help reunite those separated by conflict and natural disaster.
Imagine the following scenario. A disaster occurs in a remote part of the developing world. The local population are forced to flee their homes. Many are separated from their family and friends. With no mobile or Internet communication, finding loved ones in the aftermath of a disaster is incredibly difficult. Relief organisations go to great lengths to help people find those they are missing. Our system aims to make this process easier, faster and more secure.
We envisage a system were relief workers can to go into the remotest areas of the world with a video phone and securely record video interviews with those who have lost loved ones. These interviews are uploaded to the Reunite system where they are analysed by a global network of trusted individuals. Potential matches are reported back to relief workers, who can inform the relevant individuals and help bring them back together.
In particular, this project aims to demonstrate the feasibility of;
Machine Learning techniques will be used to ensure that the tasks carried out by the crowd are performed reliably; identifying and championing capable users and ignoring those who are less capable.
The system is being developed within the Machine Learning and Optimisation group at The University of Manchester by Peter Sutton and Lloyd Henning, under the supervision of Dr. Gavin Brown. The project is financed by the University’s EPSRC KTA scheme.