Driving by the Rules: A Benchmark for Integrating Traffic Sign Regulations into Vectorized HD Map

TL;DR

This paper tackles the gap in online HD maps by introducing MapDR, a dataset and benchmark for extracting lane-level driving rules from traffic signs and linking them to vectorized HD maps. It defines two sub-tasks—Rule Extraction and Rule-Lane Correspondence Reasoning—and provides a modular baseline (VLE-MEE) and an end-to-end approach (RuleVLM) to tackle the problem. The MapDR dataset comprises over 10,000 video clips, 400,000 front-view images, and more than 18,000 lane-level rules, enabling evaluation of rule extraction, correspondence reasoning, and their end-to-end integration with HD maps. The work demonstrates that structured multimodal fusion and vector-aware reasoning improve rule extraction and lane correspondence, offering a foundation for integrating the traffic regulation layer into online HD maps and advancing autonomous driving safety and reliability. The dataset, metrics, and baselines pave the way for future innovations in multimodal perception and HD map augmentation for driving rules.

Abstract

Ensuring adherence to traffic sign regulations is essential for both human and autonomous vehicle navigation. While current online mapping solutions often prioritize the construction of the geometric and connectivity layers of HD maps, overlooking the construction of the traffic regulation layer within HD maps. Addressing this gap, we introduce MapDR, a novel dataset designed for the extraction of Driving Rules from traffic signs and their association with vectorized, locally perceived HD Maps. MapDR features over $10,000$ annotated video clips that capture the intricate correlation between traffic sign regulations and lanes. Built upon this benchmark and the newly defined task of integrating traffic regulations into online HD maps, we provide modular and end-to-end solutions: VLE-MEE and RuleVLM, offering a strong baseline for advancing autonomous driving technology. It fills a critical gap in the integration of traffic sign rules, contributing to the development of reliable autonomous driving systems. Code is available at https://github.com/MIV-XJTU/MapDR.

Figures (16)

• Figure 1: MapDR Overview and Motivation. The image on the left depicts the comprehensive results of HD map construction for an intersection scene, whereas the image on the right illustrates the outcomes after partitioning it into three layers. Existing online mapping methods primarily emphasize the construction of the geometric and connectivity layers, neglecting the traffic regulation layer. However, precise comprehension of traffic signs and their correlation with lanes is vital for ensuring the safety of autonomous driving.
• Figure 2: Overview of the sub-tasks.$Step \ 1$$Step \ 4$ shows a case of driving by the rules. $Step \ 2$ and $Step \ 3$ demonstrates the specific role of two sub-tasks, respectively.
• Figure 3: Visualization of dataset demo. Multiple lane-level rules of a single traffic sign are annotated in $\{key:value\}$ format. Directed lines indicate the correspondence between rules and particular centerlines.
• Figure 4: Geographic location distribution of the collected traffic signs and proportions of various lane types represented in all signs. The geographic distribution is visualized based on OpenStreetMap osm.
• Figure 5: Pipeline of dataset production. The location of traffic signs are sampled from existing database then front-view images of each sign are newly collected. Vectorized map is processed in cloud server. Finally formatted rules and correspondence between rules and centerlines are annotated and organized as MapDR.