MemGeoSeg: Location-Aware Semantic Segmentation with Spatially Indexed Memory for Repetitive Driving Scenarios

Current visual perception techniques for self-driving vehicles mainly focus on the generalization across diverse scenes, and often overlook the valuable spatial consistency present in the repetitive driving routes such as public transit lines, delivery and shuttle services. In this paper we introduce MemGeoSeg, a novel multi-modal framework which enhances semantic segmentation %and object detection by exploiting scene repetitions through GPS-guided spatial priors and historical memory. Our approach introduces a hierarchical GPS embedding module, a spatially-indexed memory bank which accumulates location-specific visual knowledge and a cross-modal fusion mechanism with contrastive learning. To validate the idea of improving visual perception with repetitive driving scenarios, a new dataset, RMTD-AD, is constructed for evaluation. It contains over 13,000 annotated images across various weather and lighting conditions on repeated routes. Extensive experiments conducted on the dataset have demonstrated that MemGeoSeg significantly outperforms the state-of-the-art baseline, achieving the mIoU of 76.7% compared to SegFormer's 72.0%, with particularly strong gains in challenging scenarios like low-light and adverse weather conditions. The result shows substantial benefits of incorporating geographical contexts and historical memory for location-aware perception in intelligent vehicles. Code and datasets are available at https://github.com/Wally0924/GPS_research.git.