A Coarse-to-Fine Multi-Sensor Fusion Localization Method Based on Semantic Edge Alignment [patent]
This invention relates to the technical field of autonomous driving and visual localization. Specifically, it is a coarse-to-fine multi-sensor fusion localization method based on semantic edge alignment, comprising the steps of obtaining an HD map, raw images, a consumer-grade vehicle-mounted GPS, and wheel odometry. This invention is a low-cost, primarily vision-based high-accuracy localization method that obtains robust localization results by fusing odometry and vehicle-mounted GPS information. By extracting stable semantic features from images and performing semantic edge alignment, it estimates the vehicle pose. The alignment approach does not require explicit data association between 3D map features and 2D image features; instead, it implicitly estimates the pose by minimizing the photometric residual of the projected points.
Comments