WO2020211605A1

WO2020211605A1 - Grid map fusion method based on maximum common subgraph

Info

Publication number: WO2020211605A1
Application number: PCT/CN2020/080995
Authority: WO
Inventors: 孙荣川; 仇昌成; 郁树梅; 陈国栋; 林睿
Original assignee: 苏州大学
Priority date: 2019-04-19
Filing date: 2020-03-25
Publication date: 2020-10-22
Also published as: CN110110763A; CN110110763B

Abstract

Disclosed is a grid map fusion method based on a maximum common subgraph. The method comprises the following steps: S1, creating a grid map of an environment; S2, extracting a Harris angular point of a grid map to be fused; S3, extracting three angular points from each grid map to be fused; S4, determining whether three pairs of input angular points can form a triangle isomorphism scheme, and if not, returning to S3, and if so, performing S5; S5, iteratively constructing a polygon isomorphism scheme; S6, determining whether there is an angular point which is not substituted into the triangle isomorphism scheme, in the grid map to be fused, and if so, returning to S3, and if not, performing S7; S7, selecting an optimal polygon isomorphism scheme and a corresponding optimal transformation matrix; and S8, realizing grid map fusion according to the optimal transformation matrix and a fusion rule. The present invention can reliably realize grid map fusion and has the advantage of high fusion accuracy.

Description

A raster map fusion method based on the largest common submap

Technical field

The invention belongs to the field of mobile robot mapping, especially single-robot graph-SLAM and multi-robot SLAM, and specifically relates to a grid map fusion method based on the largest common subgraph.

Background technique

When a mobile robot is working in an unknown environment and cannot provide pose and environmental information through external devices, it is necessary for the robot to observe the surrounding environment according to its own sensors, and determine the surrounding environment and the environment based on the observed information and robot movement information. Self pose. Simultaneous positioning and map creation (simul taneous localization and mapping, SLAM) is this technology. Robots with this technology can complete more complex tasks such as navigation, path planning, and exploration. Therefore, SLAM is the key to realizing autonomy and intelligence of mobile robots.

The method based on graph optimization is currently used more SLAM algorithm, the method is divided into front-end and back-end. The front end constructs a pose map based on the sensor data, and the back end uses an optimization method to calculate the global optimal pose based on the pose map information. This method makes full use of the sparsity of the SLAM problem, that is, the feature that the data observed by the robot at different times is limited, and improves the computational efficiency of the algorithm. The front end includes sequential data association and loop closure detection. The sequential data association compares the observation information of the robot at adjacent moments. The loop closure detection compares the observation information of the robot at different times to determine the edge in the pose map. information. When using a grid map to describe the environment, the front end needs to match the sub-maps created by the robot at different times to calculate constraint information.

With the breakthroughs in theory and practical application of single-robot SLAM in recent decades, multi-robot SLAM has become one of the focuses of research in the field of mobile robots. Compared with a single robot, multiple robots can work in parallel to improve work efficiency; multiple robots can carry different sensors to provide richer observation information. Through the fusion of these observation information, more accurate maps can be obtained; multiple robots When building a map of the environment, when one or several robots fail, the remaining robots can continue to collaborate to complete the establishment of the environment model. Map fusion is one of the research focuses of multi-robot SLAM. The global map can be obtained by fusing the local environment map created by each robot.

Therefore, studying the problem of grid map fusion is of great significance for single-robot and multi-robot SLAM.

Summary of the invention

The purpose of the present invention is to provide a grid map fusion method based on the largest common sub-map, which is used to solve the problem of the front-end construction of the graph-optimized SLAM and the multi-robot SLAM map fusion.

The technical solution of the present invention is: a grid map fusion method based on the largest common sub-map, including the following steps:

Step 1. Use the ORB algorithm to extract the features of the grid maps G ₁ and G ₂ to be fused;

Step 2. Cluster the features in step 1, and calculate the cluster center;

Step 3. Use the Hamming distance to calculate the initial matching of the cluster centers, and the initial matching of the cluster centers is represented by a matrix;

Step 4. Use the backtracking method to search for three sets of initial matches that meet the constraint requirements;

Step 5. Calculate the largest common subgraph according to the location association;

Step 6. Determine whether there is an initial match that has not been accessed by the backtracking method, if yes, go back to step 4, otherwise go to step 7;

Step 7. Select the optimal largest common subgraph scheme;

Step 8. Calculate the transformation matrix according to the optimal scheme, and combine the raster map fusion strategy to achieve map fusion.

In the above, the present invention uses the ORB algorithm to extract the features of the raster map to be fused, and uses these features to characterize the raster map.

In the above technical solution, the grid map to be fused in step 1 is obtained by modeling the environment by one robot at different times or by modeling the environment by multiple robots. There are overlapping areas between.

In the above technical solution, the grid map to be fused in the step 1 is an environment description obtained by processing the internal and external sensor data of the robot by the SLAM algorithm.

In the above technical solution, the clustering center in step 2 compares the distances of the feature points, clusters those whose distance is less than the threshold value into one category, and uses the geometric center of this category of feature points as the clustering center. The descriptors of all feature points in a class are described together.

In the above technical solution, in the step 3, Hamming clustering is used to calculate the feature point descriptors, and the distance less than the set threshold is considered to meet the initial matching requirements, and then the cluster center is calculated according to the initial matching of the feature points. Initial match.

In the above technical solution, the constraint condition of the backtracking search in the step 4 is:

Among them, M ₀ ={m ₁ ,..., m _n } means to use the backtracking method to visit each row of the initial matching of cluster centers, and the initial matching of the kth row of table G ₁ and the kth cluster center in G ₂ Which of the cluster centers meet the initial matching. The proposed method assumes that the matching of cluster centers is one-to-one, so each row in the initial matching can only have one set of matches in M ₀ ;

Indicates that the i-th cluster center in G ₁ and the i′-th cluster center in G ₂ meet the initial matching requirements;

with

Respectively represent the coordinates of the cluster centers on their respective grid maps; ∈ ₀ is the error threshold; when there is only one set of matches in M ₀ , the constraint conditions are met by default.

In the above technical solution, the specific steps of the location association algorithm in step 5 are:

Step 51: When the backtracking method searches for three sets of initial matches that meet the constraint requirements, calculate the initial transformation matrix according to the matching of the three sets of cluster centers;

Step 52: According to the initial transformation matrix _, transform the cluster center in the raster map G ₂ to be fused to the coordinate system of the raster map G ₁ ;

Step 53 _: Compare the distance between the cluster center in G ₂ and the cluster center in G ₁ after transformation, and use the match whose distance is less than a certain threshold and meet the initial matching requirements as the largest common subgraph scheme of the current initial transformation matrix .

In the above technical solution, the specific steps for selecting the optimal solution in step 7 are:

Step 71: Take the one with the largest number of matching cluster centers in the largest common subgraph scheme as a candidate for the optimal scheme;

Step 72: If there is only one scheme with the largest common subgraph matching the largest number of cluster centers, then this scheme is the optimal scheme;

Step 73: If there is more than one scheme of the largest common subgraph with the largest number of matching cluster centers, then the optimal scheme is the one with the largest polygonal area formed by the matching cluster centers.

In the above technical solution, the grid map fusion strategy in step 8 is specifically:

among them,

Shows a transformation matrix transforming the grid map G ₂ grid coordinates to the coordinate system of _a grid map of G; I (·) represents the grayscale value corresponding to the coordinates corresponding to the raster; G _{12 is} represented by the corresponding grid coordinates fusion The gray value.

In the above technical solutions, the present invention can be used for the fusion of multiple grid maps.

The advantages of the present invention are:

The present invention uses ORB features to characterize the raster map to be fused; according to the ORB’s description characteristics of feature points, the Hamming distance is used to calculate the initial matching; the backtracking method is used to search for three sets of initial matches that meet the constraint requirements, and the largest Common submap; finally choose the optimal maximum common submap scheme, and calculate the optimal transformation matrix to achieve accurate fusion of raster maps.

Description of the drawings

The present invention will be further described below in conjunction with the drawings and embodiments:

Fig. 1 is a flowchart of a method according to the first embodiment of the present invention.

Fig. 2 is a grid map to be merged according to the first embodiment of the present invention.

FIG. 3 shows the ORB feature points of the grid map to be fused according to Embodiment 1 of the present invention.

FIG. 4 is the cluster center of the grid map to be merged according to the first embodiment of the present invention.

Figure 5 shows the map fusion result processed by the method of the present invention.

detailed description

Example one:

As shown in Fig. 1, the present invention provides a grid map fusion method based on the largest common submap, which includes the following steps:

Among them, the grid map is the description of the environment map obtained by processing the robot sensor information through the SLAM algorithm; the grid map to be fused is created by one robot at different times or multiple robots at the same time; the grid map to be fused has overlapping areas, See Figure 2.

As shown in Figure 3, the ‘*’ in the figure represents the features extracted from the raster map to be fused using the ORB algorithm. It can be seen from the figure that the feature points are distributed at the corner points of the grid map, which also meets the requirements of the proposed algorithm. However, it can also be seen from the figure that too many ORB features are distributed at the same corner point, which will cause repeated calculation of the algorithm and affect the calculation efficiency of the algorithm.

Step 2. Cluster the features in step 1, and calculate the cluster center;

As shown in Figure 4, the "*" in the figure represents the cluster center. The ORB feature points with similar distances are clustered into one category, and the geometric center of the feature points of this category is used to represent the cluster center. It can be seen from the figure that the cluster centers are distributed at the corners of the raster map, and there are fewer cluster centers at the same corner, which reduces the calculation amount of the following algorithm.

Because the ORB algorithm uses binary descriptors to describe the feature points, the Hamming distance is used to calculate the initial matching of the feature points, and the initial matching of the cluster centers is calculated according to the correspondence between the feature points and the cluster centers.

Among them, the constraints of backtracking search are:

Among them, M ₀ = {m ₁ ,..., m _n } means to use the backtracking method to visit each row of the initial matching of the cluster centers, and the kth row of the initial matching means the kth cluster center in G ₁ and G ₂ Which of the cluster centers meet the initial matching. The proposed method assumes that the matching of cluster centers is one-to-one, so each row in the initial matching can only have one set of matches in M ₀ at most.

Indicates that the i-th cluster center in G ₁ and the i′-th cluster center in G ₂ meet the initial matching requirements,

with

Respectively represent the coordinates of the cluster center on the respective grid map, ∈ ₀ is the error threshold. When there is only one match in M ₀ , the constraint condition is satisfied by default.

According to the backtracking method, three sets of initial matches that meet the constraints are searched. Based on these three sets of matches, the initial transformation matrix can be calculated, and the cluster centers in the raster map G ₂ to be fused are transformed to the coordinate system of the raster map G ₁ , The distance between the cluster center in G ₂ and the cluster center in G ₁ after the transformation is compared, and the cluster center whose distance is less than the threshold and meets the initial matching is taken as the largest common subgraph scheme of the current initial transformation.

Step 6. Determine whether there is an initial match that has not been visited by the backtracking method. If yes, return to step 4, otherwise go to step 7.

Step 7. Select the optimal largest common subgraph scheme;

When all the largest common subgraph schemes are calculated, the one with the largest number of matching cluster centers and the largest polygon area formed by the matching cluster centers is taken as the optimal largest common subgraph scheme.

Step 8. Calculate the transformation matrix according to the optimal scheme, and combine the grid map fusion strategy to achieve map fusion;

Wherein, the grid map fusion strategy in step 8 is specifically:

among them,

Referring to FIG. 5, it is the result of grid map fusion according to the method of the present invention. From the results, it can be seen that the method of the present invention can realize the accurate fusion of grid maps.

It should be noted that the present invention can be used for the fusion of multiple grid maps.

Of course, the above-mentioned embodiments are only to illustrate the technical concept and characteristics of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement them accordingly, and cannot limit the protection scope of the present invention. All modifications made according to the spirit of the main technical scheme of the present invention should be covered by the protection scope of the present invention.

Claims

A raster map fusion method based on the largest common submap is characterized in that it includes the following steps:

Step 1. Use the ORB algorithm to extract the features of the grid maps G 1 and G 2 to be fused;

Step 2. Cluster the features in step 1, and calculate the cluster center;

Step 3. Use the Hamming distance to calculate the initial matching of the cluster centers, and the initial matching of the cluster centers is represented by a matrix;

Step 4. Use the backtracking method to search for three sets of initial matches that meet the constraint requirements;

Step 5. Calculate the largest common subgraph according to the location association;

Step 6. Determine whether there is an initial match that has not been accessed by the backtracking method, if yes, go back to step 4, otherwise go to step 7;

Step 7. Select the optimal largest common subgraph scheme;

Step 8. Calculate the transformation matrix according to the optimal scheme, and combine the raster map fusion strategy to achieve map fusion.
The grid map fusion method based on the largest common sub-map according to claim 1, wherein the grid map to be fused in step 1 is modeled by a robot at different times or by multiple According to the robot modeling the environment, there is an overlapping area between the grid maps to be merged.
The grid map fusion method based on the largest common sub-map according to claim 1, wherein the grid map to be fused in step 1 is an environment description obtained by processing the internal and external sensor data of the robot by the SLAM algorithm .
The raster map fusion method based on the largest common sub-map according to claim 1, wherein the clustering center in step 2 compares the distances of the feature points, clusters the distances less than the threshold into one category, and The geometric center of this type of feature point is used as the cluster center, and the cluster center is described by the descriptors of all the feature points in this category.
The raster map fusion method based on the largest common submap according to claim 4, characterized in that: in the step 3, Hamming clustering is used to calculate the feature point descriptors, and the distance is less than the set threshold is considered Meet the initial matching requirements, and then calculate the initial matching of the cluster centers according to the initial matching of the feature points.
The raster map fusion method based on the largest common submap according to claim 1, wherein the constraint condition of the backtracking method search in the step 4 is:

Among them, M 0 ={M 1 ,..., m n } means to use the backtracking method to visit each row of the initial matching of the cluster center;
Indicates that the i-th cluster center in G 1 and the i′-th cluster center in G 2 meet the initial matching requirements;
with
Respectively represent the coordinates of the cluster centers on their respective grid maps; ∈ 0 is the error threshold; when there is only one set of matches in M 0 , the constraint conditions are met by default.
The grid map fusion method based on the largest common submap according to claim 1, wherein the specific steps of the location association algorithm in step 5 are:

Step 51: When the backtracking method searches for three sets of initial matches that meet the constraint requirements, calculate the initial transformation matrix according to the matching of the three sets of cluster centers;

Step 52: According to the initial transformation matrix, transform the cluster center in the raster map G 2 to be fused to the coordinate system of the raster map G 1 ;

Step 53: Compare the distance between the cluster center in G 2 and the cluster center in G 1 after the transformation, and use the match whose distance is less than a certain threshold and meet the initial matching requirements as the largest common subgraph scheme of the current initial transformation matrix .
The grid map fusion method based on the largest common submap according to claim 1, wherein the specific steps of selecting the optimal solution in the step seven are:

Step 71: Take the one with the largest number of matching cluster centers in the largest common subgraph scheme as a candidate for the optimal scheme;

Step 72: If there is only one scheme of the largest common subgraph with the largest number of matching cluster centers, then this scheme is the optimal scheme;

Step 73: If there is more than one scheme of the largest common subgraph with the largest number of matching cluster centers, then the optimal scheme is the one with the largest polygonal area formed by the matching cluster centers.
The grid map fusion method based on the largest common sub-map according to claim 1, wherein the grid map fusion strategy in step 8 is specifically:

among them,
Shows a transformation matrix transforming the grid map G 2 grid coordinates to the coordinate system of a grid map of G; I (·) represents the grayscale value corresponding to the coordinates corresponding to the raster; G 12 is represented by the corresponding grid coordinates fusion The gray value.
The grid map fusion method based on the largest common submap according to any one of claims 1 to 9, characterized in that it is used for fusion of multiple grid maps.