ADD-FAR: attacked driving dataset for forensics analysis and research

For ensuring safety and for handling emergency situations, most autonomous vehicles use remote human operators by communicating the road condition captured using LiDAR (Light Detection and Ranging) and stereo RGB cameras [10, 11]. Recent research [2, 3] has identified several possible scenarios of forgery attacks during such data communication between autonomous vehicles and human operators. Hence, there is a distinct requirement for forensics research on the multi-modal LiDAR and RGB camera data generated from autonomous vehicles. In this paper, we present a new dataset, ADD-FAR (Attacked Driving Dataset for Forensics Analysis and Research) that contains forged driving scenarios based on KITTI Vision Benchmark Suite [4, 5]. This dataset is created by identifying objects of interest using automated 3D object detection and carrying out the attacks with different levels of risk as defined in [3]. As part of the dataset, we are also providing the scripts used for automatically generating the different types of attacks. These scripts can easily be modified to work with other driving datasets apart from KITTI as well as for creating new forms of attacks.