Initialization Algorithm of Monocular Slam System Based on Unified Framework of Points and Lines

Abstract

The invention discloses a monocular SLAM system initialization algorithm based on a point-line unified framework, which includes the following steps: step 1, setting an index container with unified point-line features, so that the obtained point-line feature matching is unified in subsequent calculations F and H In the random sampling consensus algorithm of the matrix; step 2, unify the line features in the matrix calculation framework, and calculate the F and H matrices and corresponding scores in separate threads according to the midpoint of the preprocessed line features; step 3, determine the current effective matrix according to the scores , and restore the corresponding 3D point-line features based on the matrix, and then complete the initialization of the monocular SLAM system. The present invention provides an initialization method with unified point and line features, which ensures that the monocular SLAM system can make full use of image information, reduces the difficulty of system initialization, and realizes higher-precision initialization.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a monocular SLAM system initialization algorithm based on a point-line unified framework. Background technique [0002] Simultaneous localization and mapping (SLAM) has become more and more important with the development of unmanned driving and drone technology. It refers to the subject equipped with a specific sensor, without the prior information of the environment, to establish a model of the environment during the movement process, and at the same time estimate its own movement. If the camera is used as the sensor, it is called "visual SLAM". [0003] According to the type of sensor, visual SLAM can be divided into three types: monocular camera, binocular camera and depth camera. Due to the inherent "scale uncertainty" problem of monocular cameras, the SLAM system needs to complete the initialization of the system through the first two frames of image information ...

AI Technical Summary

Problems solved by technology

[0004] At present, the initialization method of the indirect method in visual SLAM is mainly to calculate the Fundamental and Homography matrices through the matching feature point pairs, determine the better matrix according to the score, and then use the 4 and 8 sets of rotation matrices R and translation vector t obtained by decomposition, and then Validate the number of valid map points that can be recovered once, and finally determine the only correct R, t matrix. However, it is difficult for this algorithm to obtain effective initialization in low-texture scenes.

[0005] In addition, in recent research, some scholars have proposed a line feature initialization algorithm based on three consecutive image frames in time. This algorithm requires 5 sets of matching line feature pairs. However, the running accuracy of the system initialized based on this algorithm is significantly lower than that of Initialization Algorithm for Point Features

[0006] To sum up, the current initialization algorithm, the single point feature algorithm has poor scene adaptability, that is, it cannot make full use of all the information in the current image frame, and the single line feature algorithm needs three consecutive frames of images to meet the "constant velocity" with high restrictions. "Assumptions, there are quite restrictive in practical application

Images