Object Perceptibility Prediction for Transmission Load Reduction in Vehicle-Infrastructure Cooperative Perception

Abstract

Vehicle-infrastructure cooperative perception is an ingenious way to eliminate environmental perception blind areas of connected and autonomous vehicles (CAVs). However, if the infrastructure transmits all environmental information to the nearby CAVs, the transmission load is so heavy that it causes a waste of network resources, such as time and bandwidth, because parts of the information are redundant for the CAVs. It is an efficient manner for the infrastructure to merely transmit the information about objects which cannot be perceived by the CAVs. Therefore, the infrastructure needs to predict whether an object is perceptible for a CAV. In this paper, a machine-leaning-based model is established to settle this problem, and a data filter is also designed to enhance the prediction accuracy in various scenarios. Based on the proposed model, the infrastructure transmits the environmental information selectively, which significantly reduces the transmission load. The experiments prove that the prediction accuracy of the model achieves up to 95%, and the transmission load is reduced by 55%.

Keywords

• connected and autonomous vehicle
• environment perception
• transmission load reduction
• object perceptibility
• machine learning