Pose estimation method suitable for monocular vision camera in dynamic environment

Abstract

The invention provides a pose estimation method suitable for a monocular vision camera in a dynamic environment. The pose estimation method comprises the steps: obtaining an image, and carrying out ORB feature point detection on each frame of image; carrying out feature point local matching on two continuous frames of images; judging whether the feature points meet constraint conditions or not, and if the feature points do not meet the constraint conditions, marking the feature points as dynamic points; performing instance-level segmentation on each frame of image to obtain an object contour,and marking all feature points in the object contour as dynamic points when the number of the dynamic points in the obtained object contour exceeds a threshold value; and performing feature matching on the feature points of the unmarked dynamic points by using an RANSAC algorithm, calculating a basic matrix of camera pose transformation, obtaining a camera rotation matrix and a translation vector,and realizing camera pose estimation. According to the invention, the pose accuracy of the camera can be improved, so that the precision of the whole SLAM system is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a pose estimation method suitable for a monocular vision camera in a dynamic environment. Background technique [0002] With the rapid development of intelligent information technology and sensor technology, smart devices such as robots have developed rapidly and are widely used in various fields of social life. Compared with traditional robots, intelligent robots have many unique characteristics, which also put forward higher requirements for intelligent robots, such as intelligent perception, autonomous decision-making, motion control, etc. The intelligentization of the robot requires the robot to be able to position itself in a complex scene, build a map of the surrounding scene, perceive the information of the surrounding scene, and complete the task independently. Simultaneous Location and Mapping (SLAM) is a basic problem in the field of computer vision and smart d...

AI Technical Summary

Problems solved by technology

[0003] In actual application scenarios, there must be moving objects other than robots in the environment. Therefore, the background in the video image sequence obtained by the robot is inconsistent with the motion of the moving objects in the environment. Direct use of video image sequences without motion segmentation will lead to visual problems. The camera pose obtained by the odometer is inaccurate, which affects the accuracy of the entire SLAM system

In the existing technology for camera pose estimation, the Faster R-CNN test network structure is difficult to segment target pixels, which in turn affects the accuracy of camera pose estimation

Images