Stereo-Aided Blockage Prediction for mmWave V2X Communications

Vision-aided blockage prediction has been recognized as a promising approach for stable Millimeter wave (mmWave) vehicle-to-everything (V2X) communications. For the supplement of channel-based prediction, the vision-aided approach can predict the communication condition in the near future, i.e., several hundred milliseconds; however, both the accuracy and prediction time of exiting work using monocular vision are not enough for mmWave V2X communications. To address the challenge, the paper proposes a stereo-aided blockage prediction that extracts explicit features for blockage prediction using a simple algorithm, i.e., stereo depth estimation. Through the emulation using a dataset generated by CARLA, we demonstrate that the proposal achieves four times faster computation than the existing method using monocular vision while improving the prediction accuracy by 15.688%.