Real-Time Quality Monitoring and Anomaly Detection for Vision Sensors in Connected and Autonomous Vehicles

The emergence of Connected and Autonomous Vehicles (CAVs) has revolutionized the transportation landscape paving the way for enhanced traffic mobility and innovative and sustainable transportation. Autonomous vehicles rely on sensor data to obtain information of the internal state of the system and the impact of the external environment to achieve self-driving autonomy. In that sense, anomalies in sensor readings may reflect faults in sensor components, which in turn can compromise the decision-making functionality of the autonomous system. Video Anomaly Detection (VAD) serves as a pivotal technology in intelligent surveillance systems, enabling the temporal or spatial identification of anomalous events within video frames. This paper proposes a supervised video anomaly assessment framework for autonomous vehicles leveraging Image Quality Assessment (IQA) and Drift Detection techniques. On this basis we adopt a two-stage approach to validate the performance of the proposed methods against a baseline Convolutional Neural Network (CNN) in a controlled low-criticality environment, as well as in more complex real-world scenarios. Our experimental results demonstrate high accuracy and robustness for anomaly detection in comparison to existing supervised techniques, which is crucial for real-time and autonomous operations of CAVs.