External point elimination method for video stream based on improved RANSAC method

Abstract

The invention discloses an external point elimination method for a video stream based on an improved RANSAC method, and the method comprises the steps: extracting feature points, tracking, and updating the tracking frequency; counting and judging the continuous tracking times of the feature points, randomly selecting minimum data capable of calculating model parameters from the classified model construction point set to perform model fitting, and calculating a basic matrix model; judging all the other feature points by using the basic matrix model, and counting internal and external point information in all the feature points; counting the number of the inner points classified in the model construction point set, comparing the number with the previous model, and updating the model; obtaining the maximum number of iterations; after one iteration is completed, adding 1 to the number of iterations; jumping out after reaching an upper limit, and outputting interior point information; marking the input feature points, if the feature points are judged to be inner points, successfully tracking and retaining the feature points, and if the feature points are judged to be outer points, tracking unsuccessfully and clearing the feature points. The precision and the operation speed of the original algorithm are improved at the same time.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to a method for removing outliers from video streams based on the improved RANSAC method. Background technique [0002] Outlier removal has always been a hot issue in the field of computer vision, and it plays an important role in various image processing problems. Taking the VSLAM system as an example, the front-end feature point method VIO will extract feature points from objects to estimate the pose of the camera, which also brings a severe challenge, that is, the matching of feature points. In the process of positioning, it is first necessary to match the feature points extracted from two adjacent frames of images, but due to the complexity of the matching problem and the large scale, it often leads to the problem of mismatching; at the same time, due to the possible Existing object motion or measurement error, even correctly matched feature points, cannot perform a goo...

AI Technical Summary

Problems solved by technology

[0004] The traditional RANSAC method has always had the disadvantages of high number of iterations, large amount of calculation and time-consuming

The growth function and similarity criteria involved in existing methods are difficult to choose for different problems, which severely limits the application and performance of this method.

During the VSLAM tracking process, the robot will pass through a lot of environments, so the data collected at different times are very different, which will lead to a large difference in the threshold of the feature points at different times, that is, N_m is difficult to determine, so the RANSAC method still needs to be applied. There are situations where it is difficult to jump out of the discriminant conditions, and the existing parameters are constantly changing in the continuous system of VSLAM, so the application effect has not been ideal all the time

At the same time, when the traditional application of RANSAC is used to remove outliers, the eight-point method is used to estimate the essential matrix. The feature of the eight-point method is the application of the least square method, which can find an essential matrix that meets most of the feature points as much as possible, which is beneficial to The stable convergence of the method, but there will be cases where the model is polluted for external points, reducing the accuracy of the model

Images