RGB-D visual-odometer method based on Census transformation and local-graph optimization

Abstract

The invention relates to an RGB-D visual-odometer method based on Census transformation and local-graph optimization. The method includes the following steps: acquiring color and depth image information of an environment through an RGB-D sensor, and using color images to calculate Census description graphs; using a direct method to carry out motion estimation on a current frame on the basis of a Census description graph, and calculating a relative pose of the current frame and a newest key frame in local graphs; for points, of which gradient information is significant but depth information islacking, in the local graphs, carrying out depth tracking estimation in the current frame; and generating a new key frame according to a pose estimation result of the current frame, inserting the sameinto the local graphs, and carrying out graph optimization and key-frame management operations on the local graphs. According to the method, the Census description graphs are used for direct-method motion estimation, optimization and management of the local graphs are combined, and real-time robustness of a visual odometer in color-image brightness changing and depth-image depth information deficiency is improved.

Description

technical field [0001] The present invention relates to computer vision technical field, more specifically, relate to a kind of RGB-D visual odometry method based on Census transformation and partial graph optimization. Background technique [0002] Visual odometry is a method of incrementally estimating the trajectory of a person or object fixedly connected to a visual sensor through an input image sequence. Compared with traditional inertial navigation and wheel odometers, visual odometers overcome the measurement error problems of inertial navigation drift and tire slippage, do not need satellite navigation, and visual sensors have the advantages of low power consumption and rich information collected, so Visual odometry has been widely concerned and gradually applied in the field of mobile robot positioning and navigation. [0003] At present, the methods of visual odometry mainly include feature point method and direct method. The main process of the feature point met...

AI Technical Summary

Problems solved by technology

The visual odometry method based on a pure monocular camera is relatively complicated. It also needs to reconstruct the three-dimensional map points while estimating the pose, and the estimated pose and three-dimensional points have no absolute scale, which is prone to scale drift and requires better initialization.

The binocular camera can obtain the depth value of the pixel in the image scene through binocular stereo matching. The workload of image processing is heavy, and it is not suitable for dark or texture features.

The RGB-D camera can acquire the color image and depth image of the scene at the same time. The depth image is acquired through the hardware structure of infrared structured light or time-of-flight method. Because it is easily affected by sunlight, it can only be used indoors, and the range of depth measurement is limited.

Images