A visual slam algorithm and system based on width learning filter

Abstract

The invention discloses a visual SLAM algorithm based on width learning filtering. First, the motion equation is established and the observation equation is transformed to obtain the state estimation problem that needs to be obtained in actual situations; Keyframes are selected from the sequence of images captured by motion and translation motion, and trained by a width learning method to predict the map of the current frame and the position of the camera. The invention also discloses a visual SLAM system based on width learning filtering, which includes a data receiving module, a motion estimation module, a matching optimization module, a width learning filtering module and a map updating module. The present invention is different from the previous method for the state estimation problem, innovatively uses the new method width learning, so as to reduce the training time without reducing the precision.

Description

technical field [0001] The present invention relates to the field of robot vision algorithms, more specifically, to a visual SLAM algorithm and system based on width learning filtering. Background technique [0002] SLAM is the abbreviation of Simultaneous Localization and Mappong, and the Chinese translation is "simultaneous positioning and map construction". It refers to the subject equipped with specific sensors, in the absence of environmental prior information, to establish a model of the environment during the movement process, and at the same time estimate its own movement. If the sensor here is mainly a camera, it is called "visual SLAM". [0003] SLAM was first proposed in 1988. At the beginning, it was used to describe the simultaneous map construction and self-localization of robots in unknown places in unknown environments. The robot establishes a map of the location environment through the environmental data acquired by the sensor, and then matches the charact...

AI Technical Summary

Problems solved by technology

[0005] In order to avoid cumulative errors, in SLAM state estimation, the usual method is to use the Kalman filter to estimate and predict the system state, but it has the disadvantages of computational complexity and sensitivity to data fusion errors, so the training time It will be relatively long, so the present invention uses a visual SLAM algorithm based on width learning filtering here. Its biggest advantage is that it can reduce the time required for training without reducing the accuracy.

Images