08. April 2020
RISC Software: “SafeSign” FFG-Funded Collaborative Project Starts
Driver assistance systems are already standard equipment in modern cars. These intelligent systems will play an even greater role in future. However, reading road signs presents the AI systems used with a challenge. RISC Software GmbH, the Johannes Kepler University Linz and highway operator ASFINAG are collaborating on a project to make these systems safer – and above all to increase the level of trust in these new technologies.
This FFG-funded cooperation is called “SafeSign”. In simple terms, it aims to improve the ability of artificial intelligence in cars to read traffic signs. That means: how do such systems in vehicles react to and recognise road signs that are damaged or dirty and therefore harder to read? To this end, everyday traffic situations like this, with which we are all familiar, must be investigated. The project’s primary objective is to give road users more confidence in the ability of intelligent systems to recognize road signs (such as temporary roadworks signs) correctly and to ensure that any automatic measures activated as a result are the right ones. All this will be enormously important in future when self-driving vehicles start sharing roads with conventional vehicles, i.e. vehicles driven by people. Along with this development, ethical principles of the engineering methods are questioned and applied, and ethically appropriate guidelines are drawn up that increase users’ confidence.
Increased confidence in intelligent systems
The rationale behind this is that humans are often not in a position to judge how well a road sign can be read by artificial intelligence. For example, a human can easily recognize and decipher an overhead LED road sign even if some of the LEDs are not working or if visibility is poor due to adverse weather. Artificial intelligence systems in self-driving vehicles, on the other hand, may not be able to do this at all or may misread the sign. Consequently, the project is developing new methods to improve AI models’ ability to interpret road signs. Deep learning methods are being applied to evaluate the recognition system on the basis of actual road sign symbols with and without defects and artificially created examples of defects.
“SafeSign” therefore aids the development of those systems that support artificial intelligence with ensuring recognition of individual signs. This would be “a road maintenance department’s third eye”, as it were, to considerably increase confidence in the functionality of such systems.
Basis of tomorrow’s mobility
The project findings, and the database of sample defects, will be published as a basis for Austrian businesses in the fields of mobility, road infrastructure and self-driving vehicles to carry out further developments. In particular it can promote the development of new traffic signs that can be read equally easily by both motorists and machines.
For infrastructure operators such as ASFINAG these are crucial steps into the future that must already be the object of research. This is the only way to ensure that developments in the field of self-driving vehicles and artificial intelligence in everyday traffic can be introduced in good time. The objectives remain high levels of safety while minimizing the number of mistakes and misjudgements of situations.
The project is funded by the Austrian Research Promotion Agency (FFG) under the "Ideen Lab 4.0” scheme.
- Duration: 1/3/2020 – 28/2/2021
- Funding scheme: Ideen Lab4.0 – 2019 call
- Project partners: Autobahnen- und Schnellstraßen-Finanzierungs-Aktiengesellschaft (ASFINAG), Johannes Kepler University Linz (JKU), RISC Software GmbH
- Grant share: Total costs EUR 141,341.00 | Grants: EUR 113,072.00 (= 80%)
Photo © RISC Software GmbH
Project team, left to right: Alexander Maletzky (RISC Software GmbH), Karl-Heinz Kastner (RISC Software GmbH), Stefan Thumfart (RISC Software GmbH), Nikolaus Kaspar (ASFINAG), Nikolaus Hofer (RISC Software GmbH), Stefan Schumann (JKU), Friedrich Robeischl (RISC Software GmbH), Karin Bruckmüller (JKU)