Visual SLAM loopback detection method based on random forest

Abstract

The present invention discloses a visual SLAM loopback detection method based on random forest. Facing visual SLAM positioning of a mobile robot in an indoor environment, the visual SLAM loopback detection method based on the random forest mainly comprises the steps of: (1) employing a regional sensing region of a convolutional neural network, performing processing of an input visual image, and obtaining a multi-dimensional feature matrix; (2) converting the feature matrix to a multi-dimensional feature column vector, inputting the multi-dimensional feature column vector into the random forestfor training, and obtaining a new feature vector; (3) employing a normalized Euclidian distance to perform similarity measurement of the feature vector obtained by visual image training of a currentframe and a feature vector of a key frame, and determining that the current frame is a loopback when the distance is smaller than a set threshold. The visual SLAM loopback detection method based on the random forest overcomes the problems that loopback detection is low in accuracy, large in calculated quantity and poor in timeliness difference by employing features set through manual work so as toimprove the position precision of the visual SLAM of the mobile robot.

Description

technical field [0001] The invention belongs to the field of robot positioning in an indoor environment, and in particular relates to a random forest-based visual SLAM loop detection method. Background technique [0002] Mobile robots perform simultaneous positioning and map construction based on visual sensor data in an indoor environment, that is, visual SLAM technology, which is the key to autonomous positioning of mobile robots. Traditional visual SLAM technology includes four parts: visual odometry, back-end optimization, loop detection and mapping. Visual odometry is mainly responsible for estimating the motion and local map between two adjacent frames of visual images, including feature extraction and image registration and other technologies. Loopback detection is mainly responsible for judging whether the robot has reached the previous position, and providing the detected loopback information to the backend for processing. The accuracy of loop closure detection di...

AI Technical Summary

Problems solved by technology

[0005] Technical problem: the technical problem to be solved by the present invention is that the traditional loopback detection method involved in the background technology is affected by environmental factors and artificial feature point selection standards, which has high error matching rate, large amount of calculation, poor real-time performance, etc. Defects, providing a visual SLAM loop closure detection method based on random forest

Images