CN109815966A

CN109815966A - An implementation method of mobile robot visual odometry based on improved SIFT algorithm

Info

Publication number: CN109815966A
Application number: CN201910139665.3A
Authority: CN
Inventors: 郑恩辉; 王谈谈; 刘政
Original assignee: China Jiliang University
Current assignee: China Jiliang University
Priority date: 2019-02-26
Filing date: 2019-02-26
Publication date: 2019-05-28

Abstract

The invention discloses a mobile robot visual odometry implementation method based on an improved SIFT algorithm. Through the depth camera installed on the mobile robot, the robot's front field of view environment is collected, and the two-dimensional image information and three-dimensional coordinate information of the space points in the front field of view environment of the mobile robot are obtained. The improved SIFT feature matching algorithm is used to process and obtain preliminary matching. The matching point pairs are obtained; the random consistency algorithm is used to eliminate the incorrect matching points in the candidate feature points, and the matching point pairs that are precisely matched are obtained; The method of the invention uses a depth camera to collect information, can directly obtain three-dimensional information of spatial points, and uses an improved SIFT feature matching algorithm to perform feature point matching, thereby significantly improving the efficiency and accuracy of mobile robot positioning.

Description

A kind of mobile robot visual odometer implementation method based on improvement SIFT algorithm

Technical field

The present invention relates to a kind of mobile robot autonomous navigation technical field, in particular to a kind of based on improving SIFT The mobile robot visual odometer implementation method of algorithm.

Background technique

In recent years, mobile robot is rapidly progressed and uses, and penetrates into the every field of life.It is traditional from In localization method, based on the method for self-locating of wheeled odometer, the drift in result is led to due to the phenomenon that there are wheel sideslips It moves, the method based on sonar and supersonic sensing interferes with each other because the two is active sensor, the positioning side based on GPS Method can not be positioned in the weak enclosed areas of signal.Due to the limitation of the relevant technologies, mobile robot is made by oneself at present Position, there are no the solutions that relative maturity is stable.

Visual odometry can be more to determine the self-positioning problem of mobile robot by acquiring and analyzing image sequence Mend traditional method for self-locating there are the problem of, enhance the efficiency and precision of localization for Mobile Robot.With inexpensive depth phase The release of machine, depth camera is more and more for localization for Mobile Robot and navigation.Different from traditional vision mileage It is more to avoid monocular odometer needs for meter, the two-dimensional image information and three-dimensional coordinate information of the available spatial point of depth camera The problem of secondary transformed coordinate system simultaneously can just solve the three-dimensional coordinate of spatial point by a large amount of calculating, improve the speed of calculating with Precision.

Summary of the invention

In order to solve the disadvantage that the prior art and deficiency, the present invention provide a kind of based on the mobile machine for improving SIFT algorithm People's visual odometry implementation method.

The purpose of the present invention is realized by the technical solution of following steps:

1) by the depth camera that is mounted in mobile robot to visual field environment in front of mobile robot into information collection, obtain The two-dimensional image information and three-dimensional coordinate information of spatial point in front of mobile robot in visual field environment；

2) two-dimensional image information obtained to step 1), using SIFT feature matching algorithm is improved, processing obtains preliminary matches Matching double points；

3) characteristic point pair of the preliminary matches obtained to step 2) rejects candidate feature using random consistency algorithm (RANSAC) Mismatching point in point obtains accurate matched matching double points；

4) kinematic parameter of mobile robot is solved using accurate matched matching double points.

The step 2) specifically: progress characteristic point detection first；Secondly the spy of binaryzation is carried out to the characteristic point of acquisition Sign point description；The characteristic point adjacent two images is slightly matched with the feature point description of binaryzation, obtains preliminary matches Matching double points.

The step 2) improve SIFT feature matching algorithm specifically includes the following steps:

2.1) it carries out characteristic point detection: two dimensional image space being established by two dimensional image, is then carried out by two dimensional image down-sampled It carries out process of convolution with gaussian kernel function again after reason to obtain the sampled images of different images size and constitute Gaussian scale-space, so Different DoG scale images is obtained by every adjacent two layers sampled images work difference in Gaussian scale-space afterwards and constitutes DoG ruler Space is spent, extreme point (Blob) is detected in DoG scale space and is used as characteristic point；And for each of DoG scale space When pixel carries out detection extreme point, each pixel with in the DoG scale image of scale 8 neighbor pixels and it is upper, Under adjacent scale DoG scale image in total 26 pixels of each 9 × 2 pixel be compared, can ensure that in this way Extreme point can be detected in scale space and two dimensional image space；

2.2) feature point description is carried out to the characteristic point of acquisition and obtains the gradient eigenvector of characteristic point, the gradient of characteristic point is special It levies vector and carries out binaryzation, obtain binaryzation gradient eigenvector, specific formula are as follows:

Wherein, a is binarization threshold, and f indicates the gradient eigenvector of characteristic point, f=[f₁,f₂...f₁₂₈], f_iIndicate that gradient is special Levy i-th of gradient information in vector, f_i∈f,b_iI-th of gradient information after indicating binaryzation；

2.3) through the above steps 2.1) -2.2) characteristic point and the binaryzation description processing for obtaining every piece image, to adjacent two Characteristic point between frame image is slightly matched, and candidate feature point is obtained: to adjacent two field pictures, using the binaryzation ladder of characteristic point Decision metric of the Euclidean distance between feature vector as characteristic point similitude in two field pictures is spent, characteristic point in two field pictures is taken Binaryzation gradient eigenvector between close a pair of of the characteristic point of Euclidean distance nearest a pair of of characteristic point and Euclidean distance time, In these two pair characteristic point, if nearest Euclidean distance is less than preset ratio threshold value divided by secondary close Euclidean distance, Europe is judged Nearest a pair of of the characteristic point of formula distance is similar, as matching double points, casts out close a pair of of the characteristic point of Euclidean distance time；

2.4) repeat the above steps 2.2) process, until all matching double points for obtaining meeting condition in two field pictures.

Data characterization calculation amount, Ji Nengbao can be greatly reduced after handling in this way by above-mentioned improvement SIFT feature matching algorithm The accuracy that match point obtains is demonstrate,proved, and can be shortened the time in matching primitives.

In the step 4), the calculation for solving the kinematic parameter of mobile robot is as follows:

4.1) following moveable robot movement parameter equation is initially set up:

4.2) then building residual sum of squares (RSS) function min { R, T } solves rotation when obtaining objective function minimum with least square method Torque battle array R and translation vector T:

Min { R, T }=| | P_qj-(RP_pj+T)||²

In formula, P_pjAnd P_qjThe corresponding three-dimensional coordinate of two characteristic points of matching double points in respectively adjacent two frame sequences image, by The three-dimensional coordinate information that the accurate matched characteristic point and step 1) that step 3) obtains obtain is combined and is obtained；Before subscript p is represented One frame image, subscript q represent a later frame image, and j is the ordinal number of accurate matched matching double points, x, y, and z represents three-dimensional coordinate；

The depth camera uses real senseD435 depth camera.

The present invention is used only real senseD435 depth camera and carries out information collection, and can directly acquire spatial point Three-dimensional coordinate information.

The SIFT algorithm used different from traditional visual odometry in the feature point description discrete consuming a large amount of time, There is a problem of real-time difference, the present invention carries out Feature Points Matching using improved SIFT feature matching algorithm, by SIFT feature Vector carries out binaryzation, to significantly improve the efficiency of localization for Mobile Robot.

Compared with the prior art, the invention has the following advantages and beneficial effects:

1, the present invention obtains the two-dimensional image information and three-dimensional coordinate information of spatial point using real senseD435 depth camera, The three-dimensional coordinate that monocular odometer needs multiple transformed coordinate system and can just solve spatial point by a large amount of calculating is avoided, is mentioned The high speed and precision that calculate.

2, the present invention proposes SIFT feature vector carrying out binaryzation, while realizing that raising guarantees detection accuracy, solves Conventional visual odometer uses SIFT algorithm in the feature point description discrete consuming plenty of time, and then leads to asking for real-time difference Topic.

Detailed description of the invention

Fig. 1 is the logical flow chart of the method for the present invention.

Specific embodiment

The invention will be further described for explanation and specific embodiment with reference to the accompanying drawing.

As shown in Figure 1, specific embodiments of the present invention are as follows:

1) by the real senseD435 depth camera that is mounted in mobile robot to visual field environment in front of mobile robot Into information collection, the two-dimensional image information and three-dimensional coordinate information of the spatial point in front of mobile robot in visual field environment are obtained；

2) two-dimensional image information obtained to step 1), using SIFT feature matching algorithm is improved, processing obtains preliminary matches Matching double points:

4) moveable robot movement parameter is solved using accurate matched matching double points:

4.1) following equation group is initially set up:

Min { R, T }=| | P_qj-(RP_pj+T)||²

In formula, P_pjAnd P_qjThe corresponding three-dimensional coordinate of two characteristic points of matching double points in respectively adjacent two frame sequences image, by The three-dimensional coordinate information that the accurate matched characteristic point and step 1) that step 3) obtains obtain is combined and is obtained；Before subscript p is represented One frame image, subscript q represent a later frame image, and j is the ordinal number of accurate matched matching double points, x, y, and z represents three-dimensional coordinate.

The two adjacent frame sequence images that the present embodiment is acquired first with depth camera, carry out altogether three groups of feature extractions with The comparative experiments of characteristic matching, time needed for obtaining former SIFT feature matching algorithm and improved algorithmic match time compare As shown in table 1.

Table 1

By 1 gained of table, improved SIFT algorithm substantially reduces matching and is consumed while realizing that raising guarantees detection accuracy Time.It solves Conventional visual odometer and consumes the plenty of time using SIFT matching algorithm, and then lead to asking for real-time difference Topic, improves the efficiency of localization for Mobile Robot.

Then the present embodiment is directed to the two adjacent frame sequence images of depth camera acquisition, solve the movement of mobile robot Parameter, the transformation results solved are following (unit: m):

T=[- 0.0351 0.0423 0.282]^T

It is recorded according to odometer, camera actual rotation angle is 6 °, and mobile is x=0.04m, y=0.04m, z=0.3m.By upper Calculation result is stated, obtains maximum absolute error as 0.018m, relative error 5%, error illustrates present invention side in allowed band Method is more accurate for localization for Mobile Robot.

More than, it is merely preferred embodiments of the present invention；But scope of protection of the present invention is not limited thereto.It is any Those familiar with the art in the technical scope disclosed by the present invention, according to the technique and scheme of the present invention and its improves Design is subject to equivalent substitution or change, should all cover in protection of the invention.

Claims

1. a mobile robot visual odometry implementation method based on improved SIFT algorithm, is characterized in that, comprises the following steps:

1) Collect information on the front view field environment of the mobile robot through the depth camera installed on the mobile robot, and obtain the two-dimensional image information and three-dimensional coordinate information of the spatial points in the front view field environment of the mobile robot;

2) to the two-dimensional image information obtained in step 1), utilize the improved SIFT feature matching algorithm to process the matching point pairs that obtain preliminary matching;

3) For the preliminary matched feature point pairs obtained in step 2), use the random consistency algorithm (RANSAC) to eliminate the incorrect matching points in the candidate feature points, and obtain the accurately matched matching point pairs;

4) The motion parameters of the mobile robot are solved by using precisely matched matching point pairs.

2. a kind of mobile robot visual odometry implementation method based on improved SIFT algorithm according to claim 1, is characterized in that: described step 2) is specially: first carry out feature point detection; The valued feature point description; the feature points between two adjacent images are roughly matched with the binarized feature point description, and the matching point pair for preliminary matching is obtained.

3. a kind of mobile robot visual odometry implementation method based on improved SIFT algorithm according to claim 1 and 2 is characterized in that: described step 2) improvement SIFT feature matching algorithm specifically comprises the following steps:

2.1) Perform feature point detection: build a two-dimensional image space from a two-dimensional image, and then perform downsampling processing on the two-dimensional image, and then perform convolution processing with a Gaussian kernel function to obtain sampled images of different image sizes and form a Gaussian scale space, Then, different DoG scale images are obtained by the difference of each adjacent two layers of sampled images in the Gaussian scale space, and the DoG scale space is formed, and extreme points (Blobs) are detected in the DoG scale space as feature points; When each pixel in the space detects the extreme point, each pixel is connected with 8 adjacent pixels in the DoG scale image of the same scale and each 9 × 2 pixels in the DoG scale image of the upper and lower adjacent scales. A total of 26 pixel points are compared;

2.2) Describe the obtained feature points to obtain the gradient feature vector of the feature point, and binarize the gradient feature vector of the feature point to obtain the binarized gradient feature vector. The specific formula is:

Among them, a is the binarization threshold, f represents the gradient feature vector of the feature point, f=[f ₁ , f ₂ ... f ₁₂₈ ], f _i represents the ith gradient information in the gradient feature vector, f _i ∈ f, b _i represent the ith gradient information after binarization;

2.3) Through the above steps 2.1)-2.2), the feature points of each image are obtained and binarized for description processing, and the feature points between two adjacent frames of images are roughly matched to obtain candidate feature points: for two adjacent frames of images , using the Euclidean distance between the binarized gradient eigenvectors of the feature points as the judging measure of the similarity of the feature points in the two frames of images, and taking the pair with the closest Euclidean distance between the binarized gradient eigenvectors of the feature points in the two frames of images Feature point and a pair of feature points with the second closest Euclidean distance. Among these two pairs of feature points, if the nearest Euclidean distance divided by the second closest Euclidean distance is less than the preset ratio threshold, the pair of feature points with the closest Euclidean distance is judged Similar, as a matching point pair, discard a pair of feature points with the next closest Euclidean distance;

2.4) Repeat the process of step 2.2) above until all matching point pairs satisfying the conditions in the two frames of images are obtained.

4. a kind of calculation mode based on the mobile robot vision mileage of improved SIFT algorithm according to claim 1 is as follows:

4.1) First, establish the following formula for the motion parameters of the mobile robot:

4.2) Then construct the residual sum of squares function min{R,T}, and use the least squares method to obtain the rotation matrix R and translation vector T when the objective function is the smallest:

min{R,T}=||P _qj -(RP _pj +T)|| ²

In the formula, P _pj and P _qj are respectively the three-dimensional coordinates corresponding to the two feature points of the matched point pair in the adjacent two frame sequence images, and the accurately matched feature points obtained in step 3) and the three-dimensional coordinate information obtained in step 1). The subscript p represents the previous frame image, the subscript q represents the next frame image, j is the ordinal number of the exact matching point pair, and x, y, and z represent the three-dimensional coordinates.

5 . The method for realizing the visual odometry of a mobile robot based on an improved SIFT algorithm according to claim 1 , wherein the depth camera adopts a real senseD435 depth camera. 6 .