BEGIN:VCALENDAR VERSION:2.0 PRODID:Data::ICal 0.24 BEGIN:VEVENT DESCRIPTION: 'Title: Securing CCTV Cameras Against Blind Spots\n When: Friday\, Aug 9\, 10:00 - 10:20 PDT\n Where: LVCC West/Floor 1/Hall 1/Tra ck 4 - [1]Map\n\n Description:\n\n In recent years\, CCTV footage has been integrated in systems to\n observe areas and detect traversing mali cious actors (e.g.\, criminals\,\n terrorists). However\, this footage h as "blind spots"\, areas where\n objects are detected with lower confide nce due to their angle/distance\n from the camera.\n\n In this talk\, we investigate a novel side effect of object detection\n in CCTV footage \; location-based confidence weakness.\n\n We demonstrate that a pedestr ian's position (distance\, angle\, height)\n in footage impacts an objec t detector's confidence.\n\n We analyze this phenomenon in four lighting conditions (lab\, morning\,\n afternoon\, night) using five object dete ctors (YOLOv3\, Faster R-CNN\,\n SSD\, DiffusionDet\, RTMDet).\n\n We then demonstrate this in footage of pedestrian traffic from three\n loca tions (Broadway\, Shibuya Crossing\, Castro Street)\, showing they\n con tain "blind spots" where pedestrians are detected with low\n confidence. This persists across various locations\, object detectors\,\n and times of day. A malicious actor could take advantage of this to\n avoid detec tion.\n\n We propose TipToe\, a novel evasion attack leveraging "blind s pots" to\n construct a minimum confidence path between two points in a\n CCTV-recorded area. We demonstrate its performance on footage of\n Br oadway\, Shibuya Crossing\, and Castro Street\, observed by YOLOv3\,\n F aster R-CNN\, SSD\, DiffusionDet\, and RTMDet.\n\n TipToe reduces max/av erage confidence by 0.10 and 0.16\, respectively\,\n on paths in Shibuya Crossing observed by YOLOv3\, with similar\n performance for other loca tions and object detectors.\n\n 1. Artificial intelligence in medicine : A comprehensive survey of\n medical doctor’s perspectives in Por tugal [2]link\, (Accessed\n 09-10-2023).\n\n 2. The impact of ar tificial intelligence along the insurance value\n chain and on the i nsurability of risks - The Geneva Papers on Risk\n and Insurance - I ssues and Practice [3]link\, (Accessed\n 09-10-2023).\n\n 3. R. Chopra and G. D. Sharma\, “Application of artificial\n intelligenc e in stock market forecasting: A critique\, review\, and\n research agenda\,” Journal of Risk and Financial Management\, vol.\n 14\, n o. 11\, 2021.[4]link\n\n 4. [B. B. Elallid\, N. Benamar\, A. S. Hafid\ , T. Rachidi\, and N.\n Mrani\, “A comprehensive survey on the app lication of deep and\n reinforcement learning approaches in autonomo us driving\,”\n Journal of King Saud University - Computer and Inf ormation\n Sciences\, vol. 34\, no. 9\, pp. 7366–7390\, 2022. (Onl ine).\n Available: [5]link\n\n 5. I. J. Goodfellow\, J. Pouget-A badie\, M. Mirza\, B. Xu\, D.\n Warde-Farley\, S. Ozair\, A. Courvil le\, and Y. Bengio\, “Generative\n adversarial networks\,” 2014. \n\n 6. I. J. Goodfellow\, J. Shlens\, and C. Szegedy\, “Explaining and\n harnessing adversarial examples\,” 2015.\n\n 7. A. Kurak in\, I. Goodfellow\, and S. Bengio\, “Adversarial examples\n in th e physical world\,” 2017.\n\n 8. A. Chakraborty\, M. Alam\, V. Dey\, A. Chattopadhyay\, and D.\n Mukhopadhyay\, “Adversarial attacks a nd defences: A survey\,”\n 2018.\n\n 9. A. Athalye\, L. Engstr om\, A. Ilyas\, and K. Kwok\, “Synthesizing\n robust adversarial e xamples\,” 2018.\n\n 10. M. Sharif\, S. Bhagavatula\, L. Bauer\, and M. K. Reiter\,\n “Accessorize to a crime: Real and stealthy attac ks on\n state-of-the-art face recognition\,” in Proceedings of the 2016\n ACM SIGSAC Conference on Computer and Communications Securit y\,\n ser. CCS ’16. New York\, NY\, USA: Association for Computing \n Machinery\, 2016\, p. 1528–1540. (Online). Available: [6]link\n \n 11. Z. Zhou\, D. Tang\, X. Wang\, W. Han\, X. Liu\, and K. Zhang\,\ n “Invisible mask: Practical attacks on face recognition with\n infrared\,” 2018.\n\n 12. S. Komkov and A. Petiushko\, “AdvHat : Real-world adversarial\n attack on ArcFace face ID system\,” in 2020 25th International\n Conference on Pattern Recognition (ICPR). IEEE\, jan 2021.\n (Online). Available: [7]link\n\n 13. B. Yin\, W. Wang\, T. Yao\, J. Guo\, Z. Kong\, S. Ding\, J. Li\, and C.\n Li u\, “Adv-makeup: A new imperceptible and transferable attack on\n face recognition\,” in Proceedings of the Thirtieth International\n Joint Conference on Artificial Intelligence\, IJCAI-21\, Z.- H.\n Zhou\, Ed. International Joint Conferences on Artificial\n Intellige nce Organization\, 8 2021\, pp. 1252–1258\, main Track.\n (Online) . Available: [8]link\n\n 14. A. Zolfi\, S. Avidan\, Y. Elovici\, and A . Shabtai\, “Adversarial\n mask: Real-world universal adversarial attack on face recognition\n model\,” 2022.\n\n 15. C. Sitawar in\, A. N. Bhagoji\, A. Mosenia\, M. Chiang\, and P.\n Mittal\, “D arts: Deceiving autonomous cars with toxic signs\,”\n 2018.\n\n 16. Y. Zhao\, H. Zhu\, R. Liang\, Q. Shen\, S. Zhang\, and K. Chen\,\n “Seeing isn’t believing: Towards more robust adversarial\n attack against real world object detectors\,”Proceedings of the\n 2019 ACM SIGSAC Conference on Computer and Communications\n Security \, 2019. (Online). Available: [9]link\n\n 17. G. Lovisotto\, H. Turner \, I. Sluganovic\, M. Strohmeier\, and I.\n Martinovic\, “SLAP: Im proving physical adversarial examples with\n Short-Lived adversarial perturbations\,” in 30th USENIX Security\n Symposium (USENIX Secu rity 21). USENIX Association\, Aug. 2021\, pp.\n 1865–1882. (Onlin e). Available: [10]link\n\n 18. T. Sato\, J. Shen\, N. Wang\, Y. Jia\, X. Lin\, and Q. A. Chen\,\n “Dirty road can attack: Security of d eep learning based\n automated lane centering under Physical-World a ttack\,” in 30th\n USENIX Security Symposium (USENIX Security 21). USENIX\n Association\, Aug. 2021\, pp. 3309–3326. (Online). Avail able: [11]link\n\n 19. W. Wang\, Y. Yao\, X. Liu\, X. Li\, P. Hao\, an d T. Zhu\, “I can see\n the light: Attacks on autonomous vehicles using invisible\n lights\,” in Proceedings of the 2021 ACM SIGSAC Conference on\n Computer and Communications Security\, ser. CCS ’2 1. New York\,\n NY\, USA: Association for Computing Machinery\, 2021 \, p.\n 1930–1944. (Online). Available: [12]link\n\n 20. S.-T. Chen\, C. Cornelius\, J. Martin\, and D. H. Chau\,\n “ShapeShifte r: Robust physical adversarial attack on faster\n r-CNN object detec tor\,” in Machine Learning and Knowledge\n Discovery in Databases. Springer International Publishing\, 2019\,\n pp. 52–68. (Online). Available: [13]link\n\n 21. K. Eykholt\, I. Evtimov\, E. Fernandes\, B. Li\, A. Rahmati\, C.\n Xiao\, A. Prakash\, T. Kohno\, and D. Song \, “Robust physical-world\n attacks on deep learning models\,” 2 018.\n\n 22. S. Thys\, W. V. Ranst\, and T. Goedemé\, “Fooling auto mated\n surveillance cameras: adversarial patches to attack person\n detection\,” 2019.\n\n 23. Z. Wu\, S.-N. Lim\, L. Davis\, and T. Goldstein\, “Making an\n invisibility cloak: Real world advers arial attacks on object\n detectors\,” 2020.\n\n 24. R. M. Oza \, A. Geisen\, and T. Wang\, “Traffic sign detection and\n recogni tion using deep learning\,” in 2021 4th International\n Conference on Artificial Intelligence for Industries (AI4I)\, 2021\,\n pp. 16 20.\n\n SpeakerBio: Jacob Shams\, Ph.D. Researcher at Cyber@Ben-Gurio n\n University\n\n Jacob Shams is a Ph.D. student at Ben-Gurion Univer sity of the Negev\n (BGU). His work addresses the security of AI models and systems\, model\n extraction attacks\, deep neural network (DNN) wat ermarking\, and\n robustness of computer vision (CV) models.\n\n Jacob is a Ph.D. researcher at Cyber@Ben-Gurion University (CBG) and\n is wor king on multiple research projects in the area of AI security.\n Jacob h olds a B.Sc. in Software Engineering from BGU and an M.Sc. in\n Software and Information Systems Engineering from BGU.\n\n '\n\n 1. #LVCCW_Lev el1_Hall1\n 2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10484446/\n 3. https://link.springer.com/article/10.1057/s41288-020-00201-7#citeas\n 4. https://www.mdpi.com/1911-8074/14/11/526\n 5. https://www.scienced irect.com/science/article/pii/S1319157822000970\n 6. https://doi.org/10. 1145/2976749.2978392\n 7. https://doi.org/10.1109%2Ficpr48806.2021.94122 36\n 8. https://doi.org/10.24963/ijcai.2021/173\n 9. https://api.seman ticscholar.org/CorpusID:207947087\n 10. https://www.usenix.org/conferenc e/usenixsecurity21/presentation/lovisotto\n 11. https://www.usenix.org/c onference/usenixsecurity21/presentation/sato\n 12. https://doi.org/10.11 45/3460120.3484766\n 13. https://doi.org/10.1007%2F978-3-030-10925-7_4\n \n\n DTEND:20240809T172000Z DTSTART:20240809T170000Z LOCATION:DC - LVCC West/Floor 1/Hall 1/Track 4 SUMMARY:Securing CCTV Cameras Against Blind Spots END:VEVENT END:VCALENDAR