Robust Place Recognition With Stereo Sequences

摘要: We propose a place recognition algorithm for simultaneous localization and mapping (SLAM) systems using stereo cameras that considers both appearance geometric information of points interest in the images. Both near far scene provide process. Hypotheses about loop closings are generated fast appearance-only technique based on bag-of-words (BoW) method. several important improvements to BoWs profit from fact that, this problem, images provided sequence. Loop closing candidates evaluated novel normalized similarity score measures context recent In cases where is not sufficiently clear, verification carried out method conditional random fields (CRFs). build CRF matching with two main novelties: use image 3-D information, we carry inference minimum spanning tree (MST), instead densely connected graph. Our results show MSTs an adequate representation additional advantages exact possible computational cost process limited. compare our system state art visual indoor outdoor data three different locations can attain at least full precision (no false positives) higher recall (fewer negatives).