Newsletter #4

Spring 2018

The Scent partners are getting ready for the pilots!

Preparation for the Scent pilots in Kifisos and the Danube Delta is in full swing. The Scent team is busy creating the time plan and planning the drone flights, the workshops and the training and dissemination material as well as the user engagement strategies to create impact in both pilots.

In the Greek pilot area, the planning is coming along smoothly. The main points of interest, the detailed pilot map and the action plan of the volunteer workshops and the campaign dates are being refined. The associated volunteer workshops are being planned in terms of their locations, the agenda, training material, duration, catering, partner presence and other aspects. The user engagement strategies are being defined in terms of the specific target groups and the channels to reach them.

As far as the Danube Delta pilot is concerned, the planning is proceeding similarly to that in the Kifisos Pilot. The time plan has been set out and the drone flights and associated arrangements, the workshops and training and dissemination material and the user engagement strategies have been planned. The points of interest of the pilot area, the pilot map and a detailed action plan of the volunteer workshops and the campaign dates are close to being finalised.

Details on how to get involved will be shared on our website and across our social media platforms soon!

The CROWD platform is the key digital infrastructure of the Scent toolkit, connecting all applications and services and orchestrating their functionality in a purposeful manner. To find out more about it, we reached out to Dr Daniele Miorandi, CEO of U-Hopper and leader of the work on the design and development of CROWD. “We see ourselves as the ‘plumbers’ of the project: we need to build the pipes, connect and maintain them so that the system as a whole keeps working correctly. Under the hood, CROWD is quite a fat boy, counting currently more than 150,000 lines of code. While this is not a quality indicator per se, and while actually only 15% has been physically written by Scent team members, it gives you an idea of the complexity of the infrastructure we have been developing.”

CROWD plays a multifold role in the big Scent picture. On the one hand, it allows local authorities to configure and manage crowdsourcing campaigns, define points of interest where pictures and measurements are needed, and monitor in real-time the pictures collected by citizens and the issues reported. On the other hand, it manages the complete lifecycle of citizen-generated images. Once a citizen takes a relevant picture using the Scent Explore app, the picture is sent to CROWD. CROWD stores a copy of the image on the cloud and makes it available to the Scent artificial intelligence ‘heart’ (the SIE), which uses advanced machine learning techniques to automatically understand what is in the image. The annotations are fed back to CROWD, which is in charge of deciding whether the ‘quality’ of the annotations is sufficient. If not, the image is made available for being further annotated, this time by people, through the CAPTCHA plugin or the Scent Collaborate app.

“One fundamental aspect of CROWD is its ability to combine artificial intelligence and human intelligence in a holistic way. When users upload a picture, they can describe what they see in the picture. We then use AI, provided by our partner IBM, to double-check whether the users are accurate or whether they miss something relevant for Scent. But if there are inconsistencies, or the machine just isn’t sufficiently sure about what’s there, we revert to humans for checking whether what the initial user and the machine said was right. This combination of artificial and collective intelligence is a distinctive factor of Scent, and something that puts us in a unique position with respect to alternative solutions,” Dr Miorandi explains.

The current prototype will be used this summer in the first wave of piloting activities. “There are still a bunch of things to be fixed, but we are confident our engineers and software guys will sort them out by the end of June, well in time for making it for the pilots in Romania and Greece“, Dr Miorandi adds.

Want to know more about CROWD? Interested in using it for your own project? Get in touch with us at https://scent-project.eu/contact-us

Scent researchers are building Deep Neural Network models for analysing the environment.

These models can automatically detect natural objects and scenes like the ones in the example below.

Analysing these kinds of natural images is very challenging, even for the modern computer vision systems.

First, the boundaries of natural scene-objects are not so well defined and quite subjective. We can for instance all point at the river bank set in stone, but can we define its location more accurately?

Second, the visual variability of these objects is huge and labelled examples are scarce. Modern vision systems use sophisticated deep neural network models requiring a large number of labelled examples to train.

In Scent, we are developing methods to achieve the same or better results while requiring significantly less examples.

Reducing the training effort for our models will also allow them to improve continuously, after they are deployed in the field, making them better and better as they are being used.

Having the ability to analyse the community photos taken in various locations over a geographical region will allow for mapping the environment in this region.

One of the cool applications of this would be feeding these maps into sophisticated geophysical models that will for example be able to predict (and help to prevent) flooding or other natural disasters.