Description
As autonomous vehicles gain mainstream popularity, the safety of these artificial intelligence-based computer vision models against interference becomes a paramount concern. This paper analyzes the vulnerability adversarial patches present to vehicular autonomous systems. Using the Ultralytics YOLO11 architecture, this study evaluates the model’s performance against a diverse set of road objects. By training the model on the Google Colab network and testing it against physically printed adversarial patches, this research analyzes the risk posed to occupants & pedestrians. The findings demonstrate significant detection failures in object detection, highlighting a critical safety gap on the roadways.
Author
Co-author
Chutima Boonthum-Denecke
(Hampton University)