SeqBEV: Bayesian Sequential Localization Using BEV LiDAR Map

We present SeqBEV, a robust and efficient framework for sequential LiDAR-based localization using a compact and structured Bird’s Eye View (BEV) map and Bayesian filtering. Traditional localization methods often struggle with feature-insufficient environments and lack explicit uncertainty modeling, leading to degraded performance. SeqBEV addresses these limitations through three key innovations: (1) a memory-efficient BEV map representation enabling real-time template matching, (2) a probabilistic observation model using normalized cross-correlation for uncertainty-aware pose estimation, and (3) a recursive Bayesian fusion framework that maintains temporal consistency. Experimental results on real-world datasets demonstrate that SeqBEV achieves superior performance compared to existing approaches, improving place recognition accuracy by 23.33% on a 50m-spaced map. The framework is highly flexible and can be extended to incorporate multi-sensor data fusion, making it suitable for autonomous driving and robotics applications.