基于改进词袋模型的自驾车辆视觉SLAM闭环检测OA

A closed-loop detection method for simultaneous localization and mapping(SLAM)of autonomous vehicles based on an improved bag-of-words(BoW)model is proposed to address the problem of increasing cumulative positioning errors during SLAM,which leads to the inability to construct globally consistent maps.The traditional BoW model is optimized by generating a vocabulary tree through the Canopy K-means clustering algorithm.A match is considered successful when the similarity between the current image and the candidate image is greater than the threshold.Double validation is performed on successfully matched images using the temporal validation method and key region covariance matrix method.The effectiveness of the proposed method is evaluated on both the public KITTI dataset and a self-collected dataset.The precision-recall curves show that compared to the original algorithm,the improved algorithm improves the recall rate by 12%while maintaining an precision of 80%.