Towards Trustworthy Perception Information Sharing on Connected and Autonomous Vehicles

Sharing perception data among autonomous vehicles is extremely useful to extending the line of sight and field of view of autonomous vehicles, which otherwise suffer from blind spots and occlusions. However, the security of using data from a random other car in making driving decisions is an issue. Without the ability of assessing the trustworthiness of received information, it will be too risky to use them for any purposes. On the other hand, when information is exchanged between vehicles, it provides a golden opportunity to quantitatively study a vehicle's trust. In this paper, we propose a trustworthy information sharing framework for connected and autonomous vehicles in which vehicles measure each other's trust using the Dirichlet-Categorical (DC) model. To increase a vehicle's capability of assessing received data's trust, we leverage the Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) model to increase the resolution of blurry images. As a result, a vehicle is able to evaluate the trustworthiness of received data that contain distant objects. Based on the KITTI dataset, we evaluate the proposed solution and discover that vehicle's trust assessment capability can be increased by 11 - 37%, using the ESRGAN model.