RESUMO
This article designs and implements a fast and high-precision multi-robot environment modeling method based on bidirectional filtering and scene identification. To solve the problem of feature tracking failure caused by large angle rotation, a bidirectional filtering mechanism is introduced to improve the error-matching elimination algorithm. A global key frame database for multiple robots is proposed based on a pretraining dictionary to convert images into a bag of words vectors. The images captured by different sub-robots are compared with the database for similarity score calculation, so as to realize fast identification and search of similar scenes. The coordinate transformation from local map to global map and the cooperative SLAM exploration of multiple robots is completed by the best matching image and the transformation matrix. The experimental results show that the proposed algorithm can effectively close the predicted trajectory of the sub-robot, thus achieving high-precision collaborative environment modeling.