Mitigating Light Detection and Ranging Spoofing Attacks in Connected Vehicle Systems Using Machine Learning-Based Anomaly Detection

Light Detection and Ranging (LiDAR) technology is one of the integral parts of systems involving connected vehicles. It provides the necessary two-dimensional accuracy to enable proper navigation, object identification, and collision avoidance through LIDAR technology. However, it is highly used and highly susceptibility to spoofing attacks: interference on signals radiated by LIDAR technology to confuse the Vehicle Perception System. This paper proposes a machine learning-based anomaly detection framework which is capable of discernment and countering of spoofing attacks. The proposed framework collects data from a set of sensors, including cameras, radar, and LiDAR, and is used to formulate a holistic, multidimensional perception model. In that sense, the system utilizes advance machine learning methods, like ensemble classifiers, neural networks, and, more importantly, autoencoders, to find anomalies in real-time that might point to spoofing activity. Simulations and empirical experiments, both benign and adversarial, evaluate the systematic approach to spoofing sophistication levels where LiDAR systems indicate noteworthy improvements in detection accuracy and robustness compared with the traditional techniques, thereby reducing false positives and negatives. In addition, through sensor fusion, the system demonstrates resilience by sustaining consistent performance even through the most complex and dynamic environments. This type of study really highlights that how the power of machine learning would be important for improvement in the cybersecurity framework of connected vehicles scalable and adaptable toward the protection of LiDAR-based systems from growing threats.