Research On Loop Closure Detection Algorithm Of Indoor Intelligent Logistics Robot Based On Feature Fusion

In the unknown environment,how to realize real-time positioning and map construction is the core problem that indoor intelligent logistics robot needs to solve.Visual simultaneous localization and mapping(VSLAM)means that the robot is equipped with a camera and other visual sensors to collect and analyze the image of the surrounding environment so as to realize real-time localization and map construction.During the operation of visual SLAM,errors will occur in the motion estimation of the vision sensor itself and the robot.If errors accumulate continuously,they will seriously affect the judgment of the intelligent logistics robot on its own positioning,resulting in the drift of the running track,and even the failure of drawing construction.By comparing the current scene and the historical scene,the loop closure detection can effectively eliminate the above cumulative errors,so that the intelligent logistics robot can accurately complete the SLAM work.However,in the real scene,there are many factors that will lead to changes in the visual appearance of the environment and affect the accuracy of loop closure detection.This paper fully analyzes the shortcomings of the existing loop closure detection methods and proposes a loop closure detection algorithm for indoor intelligent logistics robot based on feature fusion.The main work and achievements are as follows:(1)In the aspect of feature extraction and matching,the accuracy of the traditional method is low,and it is easy to mismatch,and a single feature is difficult to fully and accurately describe the whole image.To solve this problem,the deep neural network model is used in this paper to conduct semantic segmentation of images.According to the mask blocks in the segmentation results,the noise is removed and three-dimensional semantic features and appearance features are obtained.The semantic appearance similarity score of images is calculated to compare the similarity between images.Experimental results show that the proposed method can effectively improve the accuracy of feature matching and loop closure detection.(2)In the aspect of loop closure detection,the traditional algorithm often adopts a single type feature,which is difficult to describe the whole image comprehensively and accurately,and can not reasonably calculate the image similarity.In this paper,a variety of features contained in the image are extracted,an image similarity calculation model is constructed integrating multi-dimensional visual features,and a multi-layer screening mechanism is established,so as to narrow the image comparison range,reduce the calculation difficulty and comparison times,and shorten the overall running time of the algorithm.At the same time,in order to ensure the authenticity of the loop closure and avoid false positive loop closure,Siam network was introduced to verify the detected loop closure and determine the final loop closure.Through the experiments on the public data set and the self-shot dataset,it is proved that the proposed loop closure detection algorithm has been improved in accuracy and speed.