Efficient Cross-Country Data Acquisition Strategy for ADAS via Street-View Imagery

TL;DR

The paper tackles cross-country ADAS/ADS deployment by addressing domain shifts in legislation, infrastructure, and visual cues through a street-view–guided data acquisition strategy. It introduces two foundation-model–based POI scoring methods—a KNN feature-distance score and a visual-attribute score using visual attribution—to identify informative locations for target-domain finetuning, evaluated under a collect-detect proxy protocol with co-located driving and Mapillary street-view data. Experimental results in traffic sign detection show that the visual-attribute method achieves comparable improvements to full data with only a quarter to half of the target data and exhibits better stability across countries, while the feature-distance method is more sensitive to viewpoint variability. A practical cost analysis demonstrates that large-scale street-view processing is economically feasible, underscoring the approach’s potential for scalable, data-efficient cross-country adaptation in real-world ADAS deployments.

Abstract

Deploying ADAS and ADS across countries remains challenging due to differences in legislation, traffic infrastructure, and visual conventions, which introduce domain shifts that degrade perception performance. Traditional cross-country data collection relies on extensive on-road driving, making it costly and inefficient to identify representative locations. To address this, we propose a street-view-guided data acquisition strategy that leverages publicly available imagery to identify places of interest (POI). Two POI scoring methods are introduced: a KNN-based feature distance approach using a vision foundation model, and a visual-attribution approach using a vision-language model. To enable repeatable evaluation, we adopt a collect-detect protocol and construct a co-located dataset by pairing the Zenseact Open Dataset with Mapillary street-view images. Experiments on traffic sign detection, a task particularly sensitive to cross-country variations in sign appearance, show that our approach achieves performance comparable to random sampling while using only half of the target-domain data. We further provide cost estimations for full-country analysis, demonstrating that large-scale street-view processing remains economically feasible. These results highlight the potential of street-view-guided data acquisition for efficient and cost-effective cross-country model adaptation.

Figures (8)

• Figure 1: Traffic signs exhibit strong domain shifts across countries. In Poland, warning signs have yellow backgrounds, whereas in Sweden, generic warning signs use a vertical line instead of an exclamation mark alibeigi2023zenseact.
• Figure 2: Workflow of the street-view–guided cross-country data acquisition strategy. In this example, Germany serves as the source country providing training data, while Poland is the target deployment country. Instead of using data from the entire target region, the model is finetuned or validated using only the selected , enabling more efficient deployment.
• Figure 3: Two proposed foundation model-based methods to detect .
• Figure 4: Prompt used for traffic sign attribute extraction.
• Figure 5: Examples of co-located images. (a) shows the on-board image from the alibeigi2023zenseact dataset, while (b)-(d) are street-view images within a 10 m radius from Mapillary mapillary_api_docs.