CrowdEvaluator: A Data-Driven approach for quality evaluation in crowdsourced HD maps production

• A data-driven approach is proposed for reference-free quality evaluation in crowdsourced high-definition map production. • Discriminative temporal shapelets from vehicle-end time-series data assess measurement reliability. • A Markov Random Field–based consistency module enforces local spatial coherence in the predicted quality labels. • The framework supports scalable automated quality control for HD map production. Crowdsourcing provides an efficient pathway for large-scale High-definition (HD) maps production. Quality evaluation in crowdsourced HD map production remains a significant challenge. Existing quality evaluation methods typically rely on ground truth or manual inspection. Such dependence renders these methods incapable of supporting automated quality control in the crowdsourced map production process. To address this issue, we propose a data-driven approach that exploits the spatiotemporal patterns of vehicle-mounted sensor observations to strengthen quality control for HD map features. A multivariate time series classifier is developed to evaluate the quality of local map elements. The classifier models the spatiotemporal patterns of the data collection process using shapelet-based representations together with generic feature encodings. A post-processing strategy is applied to ensure globally consistent and spatially smooth predictions. Experimental results demonstrate that the proposed method is effective across typical road scenarios. It achieves a classification accuracy of 89.5%. Our method offers a practical approach for quality control in crowdsourced HD map construction. It contributes to improving production efficiency and positional accuracy in crowdsourced mapping workflows.