Physical Attack for Stereo Matching
Yang Liu, Jucai Zhai, Chihao Ma, Pengcheng Zeng, Xin’an Wang, Yong Jun Zhao · 2024
Stereo matching networks have received extensive attention in the field of autonomous driving due to their reliability and low cost in depth estimation tasks. This is largely due to the development of deep stereo networks. However, researchers have found that deep neural networks are vulnerable to adversarial attacks, and their security is worrying. So far, physical world attacks against deep stereo networks have not been systematically studied. Therefore, we extend the physical world attack to the domain of stereo matching. We design a new patch attack method for stereo matching, the disparity map attack. Extensive experiments show that our method outperforms previously published attack methods. Our patch poses a certain threat to stereo matching networks in real-world attacks.