Laser Vision for Low-Coupling Indoor Positioning

We proposed a low-coupling indoor localization algorithm for mobile robot based on laser and vision sensors with better robustness and real-time performance. Firstly, mobile robot kidnap problem was decomposed into pre-kidnap state judgment and post-kidnap localization recovery. Mobile robot was jointly judged by a probabilistic model and CLIFF to determine whether abduction occurs or not. The post-kidnap localization recovery problem was solved by using KLD resampling based on coarse localization. The ORB feature descriptors were extracted from the images of different frames and the corresponding pose information was recorded. Then the VP DBOW was constructed, followed by filling the particles with the pose information of coarse localization during resampling or the first global localization. Coarse localization was implemented to match the current frame in V-PDBOW to get the pose information corresponding to the a posteriori observation frame, which could reduce the iteration time of the algorithm and solve the recovery of localization after abduction. Finally, the likelihood domain model was optimized and a high-weight laser beam was adopted as the calculation basis for evaluating the particle weights to avoid the influence of noisy data on the evaluation, thus improving the robustness of the localization algorithm and more rapidly correcting the positional distribution to the true distribution.