| The technology of mobile robot accurately locating its position by collecting images from sensors in unknown environment and reconstructing the map of the surrounding environment is called simultaneous localization and mapping(SLAM).As the front end of slam technology,visual odometer mainly extracts and matches the features of the images extracted by the sensor,so as to estimate the motion trajectory of the robot and provide data support for the follow-up work.In this paper,the front-end visual odometer of mobile robot is studied.The main work is as follows:This paper studies the sensor information reading in mobile robot,and introduces the significance of camera calibration,the conversion relationship between four coordinate systems,and the mainstream camera calibration methods.Finally,choose to use camera under ROS operating system＿The calibration function package completes the calibration of the selected inter realsense d435 i camera and obtains the internal and external parameters of the camera.Then,this paper deeply analyzes the complexity of logic judgment of fast feature point extraction algorithm,and proposes an improved fast4-4 extraction algorithm.The algorithm calculates the gray level of pixels in the vertical and horizontal directions on the circumference with radius 3,and compares it with the gray level of pixels in the center of the circle.Only when the threshold condition is met can it be included in the candidate corner.The calibrated rgb-d camera is used to continuously sample various actual scenes,and the obtained images are used as samples.After comparing the extraction speed and number of feature points before and after the improvement of the algorithm,it is found that the average time per image of fast4-4 algorithm is about 5 milliseconds,which is 1 ～ 2 milliseconds shorter than the traditional algorithm,and the number of feature points is more.Experiments show that the improved fast4-4 algorithm can effectively improve the efficiency of feature point extraction.Then,aiming at the problems of feature points clustering and nonuniformity in the image,Harris corner scoring algorithm combined with non maximum suppression is used to screen the feature points,and only 3 points are retained × The feature point with the highest score in the 3-pixel area.After the feature point screening is completed,according to the different number of feature points in the image,the fast approximate nearest neighbor algorithm or violent matching algorithm is selected for feature point matching,and then the Hamming distance false matching screening algorithm is used to eliminate the false matching point pairs to obtain the matching results.Finally,the ICP algorithm and Ba optimization algorithm are used to estimate the pose of the matched standard dataset image.The calculated pose and standard pose are drawn with the trajectory drawing tool Evo,and the error is compared to demonstrate the feasibility of the improved feature point extraction and matching algorithm. |
