Would you consider the projection of the gentleman real? Well, as shown on the right corner of the picture, the Telsa Model X would.
Would you speed down the highway after seeing this speed sign projected on the tree? I suppose you probably won’t. Surprisingly, as shown on the left side of the picture, the Tesla Model X would.
Do those facts pose a problem or any kind of danger? They surely do.
“Phantom” objects and signs
A phantom is defined as a depthless visual object. Even though this object doesn’t have a 3D dimension (they don’t pose a threat to the driver), a “better safe than sorry” policy of the assistance system considers them to be real. This leads to applying the brakes immediately and causing havoc on the road. A phantom object can be created by a projector or be presented via a digital screen (e.g., billboard). The signs (speed sign, stop sign, etc.) are being recognized, even if they appear for a few milliseconds – split-second phantoms.
Split-Second Phantom Attack
Recently, the gap in advanced driver-assistance systems has been used by hackers. Cyber attacks can be applied through advertisements and cause ADAS to trigger a reaction. Attackers use an algorithm to hide a phantom in a random advertisement. Phantom attacks are really tricky. They can be applied remotely and do not leave any evidence at the attack scene. Moreover, they don’t require any special expertise, complex preparation, or expensive equipment.
Luckily, researchers are one step ahead. They came up with a detector that doesn’t require any additional changes to be made to the existing infrastructure. “It consists of four lightweight deep CNNs that assess the realism and authenticity of an object by examining the object’s reflected light, context, surface, and depth” a research paper from Ben-Gurion University of the Negevsay states. The ongoing research might limit the opportunities for attackers to apply phantom attacks but won’t eliminate them.