Visual odometer implementation method based on ORB feature points and GMS matched filter

Abstract

The invention discloses a visual odometer implementation method based on ORB feature points and a GMS matched filter. The visual odometer implementation method comprises the steps: employing an RGBD camera or a binocular camera for collecting video streams; for each frame of image, based on the current pose of the camera, obtaining feature points (including two-dimensional pixel coordinates, three-dimensional world coordinates and a brief descriptor of angular points corresponding to the feature points) in the image; matching the feature points between the current frame and the previous framein combination with a GMS matching filter; solving a PnP problem according to the matched feature points of the previous frame to obtain a camera pose of the current frame; simultaneously optimizing the camera pose and the three-dimensional world coordinates of the feature points of a plurality of nearest continuous frames by using a Bundle AdJustment; calculating a bag-of-word vector of the current frame by using a bag-of-word model, and judging whether the current frame needs to be stored as a key frame or whether loopback occurs or not; if loopback occurs, optimizing the related key framesby using fast loopback optimization. According to the method, under the condition of ensuring that the re-projection error of the camera is small enough, compared with a traditional singular value decomposition method, operation is simpler and more convenient, and real-time performance is ensured.

Description

technical field [0001] The invention belongs to the field of autonomous navigation of robots, and relates to a method for realizing a visual odometer based on ORB feature points and GMS matching filters. Background technique [0002] Odometer is a technology that uses sensors to estimate the pose of the object being measured. At present, there is an increasingly urgent demand for high-precision and robust odometers in all walks of life, especially in the fields of mining, military, and exploration. Acquisition of environmental information and effective pose estimation in situations such as three-way positioning. Visual odometry is an algorithm for pose estimation based on visual information. Compared with the currently commonly used laser odometer and inertial navigation odometer, the visual odometer has a lower cost, but the corresponding implementation method is not mature enough. In relatively complex scenarios, its robustness, real-time performance and accuracy cannot...

AI Technical Summary

Problems solved by technology

Their test results on public data sets such as KITTI and EuRoC are acceptable, and the accuracy is high, but the overall performance in the actual application process is poor, and it is difficult to overcome the interference problem in complex scenarios. The so-called "real-time" is also Based on a computer with strong performance, and the detection of ORB (a fast feature point extraction and description algorithm) feature point detection is more computationally intensive and easy to lose, etc.

Images