Closed Loop Detection Based Vision SLAM For Mobile Robot

The closed-loop detection problem of robot is a important basis when the robot in positioning and navigation.But due to the cumulative error of the visual sensor, it can’t judgment the closed-loop response. Simultaneous localization and mapping has the problems of low estimate accuracy and long time consuming when mobile robot faced with the closed loop detection, To deal with these problems, The major research contents are as follows:Firstly, this paper summarizes the background and significance of the research, and analyzes the status and existing problems of the vision-based mobile robot closed-loop detection algorithms in detail, and establishes a variety of mobile robot system models,these systems model are the basis of a platform for the vision-based mobile robot localization problem.Secondly, this paper aiming at the effects of fuzzy image feature points extraction,using the algorithm of maximum expected to blind to blurred image, image restoration by maximum likelihood estimation to the clear picture, then use the SIFT operator, the MATLAB simulation experiment proves that the use of blind to blur the image improves the accuracy of image feature matching.Again, in view of the closed-loop detection problems such as low accuracy, long time consuming in the algorithm, this paper design a closed-loop detection algorithm based on improved visual dictionary,and put forward an algorithm of building visual dictionary based on quadratic notation, Converts to extract the continuous image characteristics of discrete visual words, Through the simulation experiments of scene classification, The improved algorithm improves the detection accuracy, shorten the calculation time.Finally, in view of the likelihood function in the naive Bayes algorithm to estimate the accuracy of the observation model and algorithm of redundancy, long computing time,designed the improved Bayes algorithm model, and combined with improved BOVW model, the algorithm can be divided into the stage of training and predicting stage, achieve the Bayesian framework of image similarity computation. Solved using traditional Bayesian probability model of likelihood estimation error of the problems of uncertainty,the algorithm performance contrast experiment, verify the performance of the improved algorithm.