Learn about critical vulnerabilities in CCTV surveillance systems through a DEF CON conference talk that explores how object detection confidence varies based on camera positioning and environmental factors. Dive into comprehensive research examining location-based confidence weaknesses across multiple object detection systems including YOLOv3, Faster R-CNN, SSD, DiffusionDet, and RTMDet under various lighting conditions. Understand how these "blind spots" - areas where objects are detected with lower confidence due to angle and distance from cameras - persist across different locations and times of day. Discover TipToe, a novel evasion attack that exploits these weaknesses to construct minimum confidence paths between points in CCTV-monitored areas. Examine real-world demonstrations using footage from Broadway, Shibuya Crossing, and Castro Street, showing how TipToe can reduce detection confidence by significant margins. Gain valuable insights into improving CCTV system security by understanding these inherent limitations in current object detection implementations.
Overview
Syllabus
DEF CON 32 - Securing CCTV Cameras Against Blind Spots - Jacob Shams
Taught by
DEFCONConference
Reviews
5.0 rating, based on 1 Class Central review
Showing Class Central Sort
-
The class was great, I fully understanded but in some cases we need more explainations it would help us to improve our knowledge thank you sir