Business Accelerator
Smart Applications of Artificial Intelligence
AI is frequently used as a catch-all buzzword for many different things. Sometimes this is justified – smart software applications really are penetrating further into our digital lives, and improving them, too. We demonstrate with our work that artificial intelligence, for us, is one of many intelligent tools. However, it shouldn’t be the entire purpose of an application.
We find projects exciting when users don’t immediately recognize the AI. When the focus is on great experiences instead.
The Bikespector app provides users with a big-data analysis that predicts bike-sharing availability at a chosen location on maps of Cologne, Düsseldorf, Berlin, and Frankfurt. For the Web application, we analyzed different sources of data, such as the weather forecast and events calendar. By constantly matching data with the nextbike API, we were able to obtain the current locations of the shared bikes and, what’s more, gather these locations in a database over the course of one year. Using that database, we trained an algorithm to find out the connections between individual elements of data. Does temperature influence bike-share hires, for example, or is it the weather in general? Based on this data, the smart software created a mathematic formula – the machine-learning aspect of this project.
Even a seemingly trivial task such as this can benefit from applications of AI. Visualizing the map ended up requiring more effort than developing the algorithm itself.
For instance, a user can select a location on the Bikespector map or type in a street name. We add the next day’s weather and the algorithm then calculates a degree of probability. It can predict how many bikes will be available at a given location at a given time – the data-prediction aspect.
In other applications, AI can play a key role, for example when tracking and analyzing images. Azure, Microsoft’s AI-service platform in the cloud, has a facial-recognition function, for example. Microsoft hired denkwerk to bring this field of knowledge to life for users. To do this, we developed an installation for the 2018 “Year of the Rooster” conference in Munich. It involved attendees interacting with a projection screen – they saw themselves in a video stream that featured funny and individual illustrations in place of their heads. How did we do this? Uploading the entire video stream to the Microsoft cloud server would have taken up too much bandwidth. That’s why our local algorithm identified the face of each attendee in front of the wall using various markers and their relation to each other.
For the facial-recognition algorithm, we worked with OpenCV, a code library for applications with real-time computer vision. Microsoft’s cloud AI used the face sent by us to recognize age and gender, eyewear and facial hair, and even certain emotions, too. It sent this information back to us for the projection screen. We assigned an identifier to the individual characteristics and composed a collage-like illustration from them – an individual face with recognition effect. The result was displayed in the real-time video stream. If visitors walked past an hour later, their individual illustration would be displayed again using the assigned identifier.
For the Wallraf Richartz Museum in Cologne, we developed a variation of facial recognition, namely, an algorithm to recognize images. The challenge we faced with this was the same experienced by any traditional museum: getting young people excited about exhibitions. That’s why denkwerk brought art into the public sphere – and to young people – for the show celebrating Italian painter Tintoretto in the fall and winter of 2017/2018.
We developed the Tintoretto2Go app. Using it, users could explore the works of the painter on their smartphone throughout the entire city. The app displayed augmented-reality content about individual paintings when the smartphone’s camera was held in front of various advertising displays, like at a bus stop, for example.
Six different advertisements triggered the AR application. However, we deliberately decided not to use QR codes to trigger it, as is normally done for AR. (We would’ve had to put the QR codes on all the posters in the city for this). Instead, we trained an algorithm to recognize individual advertisement posters. It even recognized posters when they were obscured by passers-by. We used the Vuforia software platform to realize the app and an image-recognition code for the poster subject. Since the city’s advertisements were swapped over every two weeks, we also trained the algorithm for new advertisement subjects using photographs.
AI can be an extremely fitting match in a creative context, and we showed proof of that with the banner we designed for the 2019 ADC Digital Experience Conference. We created it using an illustration that was transformed by AI – and then processed in turn by an illustrator. To go about this, we used an algorithm that digitally edited the illustration using deep learning in order to adapt it to a specific style. This type of algorithm is called neural style transfer.
What Ardi lacks in creativity, it makes up for in being a highly useful work tool – created by us. The research tool groups together search results using Microsoft Azure Cognitive Services and presents them in a condensed overview. We used text analysis and a content-specific algorithm for this. Our prototype won second place for the AI Award Agency Edition from the Microsoft Partner Network Germany.
In the end, though, this is far from everything that denkwerk has to offer and achieve in terms of AI. We want to go further and develop and shape innovation, together with our clients. Get in touch with us – we’d love to talk with you!
business@denkwerk.com
