RevPv8: Reverse Information for Road Space and Lane Line Segmentation in Highway Surveillance Scene

Abstract

Road space and lane line segmentation are vital tasks in intelligent traffic surveillance, yet they receive less attention compared to similar tasks in autonomous driving. In highway surveillance, segmenting occluded road areas and lane lines is crucial for comprehensive perception, though it adds complexity. While recent advances in multi-task learning have shown promise, many models fail to fully explore task interdependence. In this paper, we introduce RevPv8, a novel benchmark model with a dual-column decoder: the Road Column and the Lane-line Column. To improve information sharing within the decoder and fully exploit task relevance, we incorporate a multi-level information sharing mechanism, reverse information interaction, and auxiliary supervision tasks. This enables road space features to guide lane line segmentation during decoding. Additionally, we present a dataset Highway10K, designed for road space and lane line segmentation in highway scenarios. Experimental results on Highway10K and BDD100K show that RevPv8 surpasses other multi-task models, particularly in lane segmentation, providing more accurate highway perception for intelligent traffic surveillance.