Towards an AI-Based After-Collision Forensic Analysis Protocol for Autonomous Vehicles

Abstract

Safety-critical applications in the cooperative vehicular networks are built to improve safety, traffic efficiency and handle emergencies by communicating the road condition captured using data from sensors (camera, LiDAR, RADAR, etc.). These cyber-physical systems maintain records of the data received from its sensors to make decisions while driving on road. Such proliferation of data opens possibilities of scenarios where attackers can forge into the system with unrestricted access to the internal network of the vehicle and perform malicious acts. Due to the possibility of such acts, it is crucial how forensic analysis should be carried out in case of traffic accidents that include autonomous vehicles (AV). In this paper, we propose a forensic investigation protocol on autonomous vehicles, specifically to investigate if there was an attack that targeted the vehicle sensors. The proposed process consists of three main phases: data curation, analysis and decision making. We argue that, by using supervised deep neural network-based architecture YOLO trained in the Darknet framework and tested with SORT, an effective model to detect traffic data can be built to perform forensic investigations.