Self-driving cars are gradually becoming part of our transportation system, and their systems are constantly being refined so that they can anticipate situations much like humans do. Recently, however, a major problem has been uncovered that these cars are unable to handle, and which could actually cause an accident. Stickers, graffiti, or other images on traffic signs pose major complications for the systems, and the cars are unable to “read” the signs correctly.
A great deal of research and testing is currently being conducted in the field of autonomous vehicles. These vehicles must be absolutely safe and should be able to respond to all situations they may encounter while on the road. Cars must correctly “read” road signs and respond appropriately.
However, a recent study has shown that in many cases, cars misread signs if they have stickers or other symbols on them that do not belong there. The system works by comparing the sign to a database of images of identical traffic signs. However, a sticker can confuse the system to such an extent that it identifies a STOP sign as a 45 km/h speed limit. This occurred in as many as 73% of cases during testing. Other signs were also tested; for example, the system misidentified a “Turn Right” sign as a “STOP” sign after graffiti was added.
However, this is not the only factor that caused the system to become confused. Another problem was faded or otherwise improperly manufactured signs. A sign indicating a right-hand curve was tested; its arrow was faded, and in two-thirds of cases, the cars interpreted this sign as a 45 km/h speed limit or some other speed.
Experts fear that such problems could become an opportunity for cyberattacks in the future. Although this bears very little resemblance to typical cyberattacks, we include it here because it is possible to influence the functioning of a specific system. In this way, an attacker could cause a truly serious traffic accident and thus endanger people’s lives.
Researchers have come up with a solution. They designed a system to prevent incorrect sign recognition. Currently, cars rely solely on a database of traffic signs and search for matches based on that database. However, it would be advisable for the car to also take the actual road situation into account. For example, when entering an intersection, there is certainly no “speed limit increase” sign there; rather, the car can expect to encounter a STOP sign and very similar situations. In the future, it will be essential to identify such problems and come up with solutions before these vehicles are fully integrated into traffic.
SOURCES:
https://www.youtube.com/watch?v=0JVvr9czIqg
