SLAM Of Indoor Mobile Robot Based On Depth Camera

Simultaneous Localization and Mapping(SLAM)technology,due to its small size,high efficiency,low power consumption,high accuracy,and portability,has become more and more widely used in industrial and civil fields,and it also has a broad application prospect.This paper applies this technology to the simultaneous positioning and mapping of indoor mobile robots,and introduces the related theories and implementation methods of simultaneous positioning and mapping.This article uses RGB-D cameras RealSense to build and test SLAM systems.The specific experiments are as follows:Firstly,the principle of camera imaging model is introduced.The depth camera is calibrated to obtain the internal and external parameters of the camera,which provides more accurate image information for subsequent feature extraction and registration,pose estimation,and so on.Secondly,this paper conducts feature extraction and matching experiments on images.According to the motion characteristics of a plane-moving wheeled robot equipped with an RGB-D camera,a method of screening for mismatched feature point pairs based on equal height constraints and linear constraints is proposed.In this paper,under the constraints of the camera motion of the planar motion robot,the equal height constraint and the straight line constraint of the matching feature point pairs are established,and then this condition is used to filter out the mismatched point pairs generated by the brute force matching method and obtain high-quality frames Match point pairs.Finally,this paper combines the mismatching feature point screening algorithm with a random sampling consensus algorithm to estimate the motion of the robot camera.Experiments show that compared with the original random sampling consistency algorithm,the feature point screening algorithm proposed in this paper can effectively improve the efficiency of the random sampling consistency algorithm and improve its calculation accuracy.In the pose estimation process of the camera,the method of graph optimization is used for optimization.Then,a bag-of-words model based on visual features is used for loop detection,a dictionary of key frames is trained and the similarity is calculated,which improves the system's loop detection speed and accuracy.Finally,the algorithm was verified with internationally published data sets,and experiments were performed in the laboratory.The experimental results show that the algorithm can accurately locate and map the robot.