阅读说明：本技术 基于语义的扫地机器人自适应清扫方法 (Semantic-based self-adaptive sweeping method for sweeping robot ) 是由 张希 于 2020-05-28 设计创作，主要内容包括：本发明涉及一种基于语义的扫地机器人自适应清扫方法,通过获取待清扫场景的全局地图,并且利用不同的语音对全局地图中的不同物体种类做标识,然后再将对应不同语义的数值填入到对应物体种类的栅格中,利用分类聚合的方法对具有相同语义的栅格做分类聚合处理,且按照栅格轮廓区块划分各分类聚合后的区域,从而使得扫地机器人可以自动地针对目标全局地图内的各不同语义清扫区域分别执行对应的清扫工作模式,实现了扫地机器人可以自动地针对不同清扫区域做识别以及清扫工作模式调整,提高了扫地机器人的智能化自动清扫效率。(The invention relates to a semantic-based sweeping robot self-adaptive sweeping method, which comprises the steps of obtaining a global map of a scene to be swept, identifying different object types in the global map by using different voices, filling numerical values corresponding to different semantics into grids corresponding to the object types, classifying and aggregating grids with the same semantics by using a classification and aggregation method, and dividing each classified and aggregated area according to a grid outline block, so that a sweeping robot can automatically execute corresponding sweeping work modes aiming at different semantic sweeping areas in a target global map, the sweeping robot can automatically recognize different sweeping areas and adjust the sweeping work modes, and the intelligent automatic sweeping efficiency of the sweeping robot is improved.)

1. A self-adaptive sweeping method of a sweeping robot based on semantics is characterized by comprising the following steps:

step 1, acquiring a global map of a scene area to be cleaned; wherein the global map is a rasterized map;

step 2, identifying and obtaining different object types in the global map;

step 3, respectively identifying various object types in the global map by using different semantics to obtain a semantization global map subjected to semantization identification processing;

step 4, corresponding numerical values are given to different semantics, the numerical values corresponding to the semantics are respectively filled into grids corresponding to the object types in the semantic global map, and the numerical global map after numerical value filling processing is obtained;

step 5, performing classification and aggregation processing on the regions with the same semantics in the numerical global map respectively to obtain a classified and aggregated global map; the regions with the same semantics in the aggregated global map are regions of the same type;

step 6, carrying out block division on grids in the same type of regions in the aggregated global map according to grid outlines to form a global map with different semantic cleaning regions, and taking the global map with different semantic cleaning regions as a target global map;

and 7, respectively executing corresponding cleaning working modes by the cleaning robot aiming at different semantic cleaning areas in the target global map.

2. The semantic-based sweeping robot adaptive sweeping method according to claim 1, characterized in that in step 7, the sweeping robot performs different sweeping modes of operation for different semantic sweeping areas within the target global map.

3. The semantic-based sweeping robot adaptive sweeping method according to claim 1, wherein the scene area to be swept is at least one of a living room, a bedroom, a kitchen, a dining room and a bathroom.

4. The self-adaptive sweeping robot sweeping method based on the semantics as claimed in any one of claims 1 to 3, wherein in step 7, an optimal sweeping path is planned for each different semantic sweeping area in the target global map, and the sweeping robot sweeps the corresponding semantic sweeping area according to each optimal sweeping path.

5. The self-adaptive sweeping method for the semantic-based sweeping robot according to claim 1 or 2, wherein the step 5 is to obtain the aggregated global map according to the following steps a 1-a 4:

a1, acquiring the total number of semantic large classes in the semantic global map; the total quantity of the obtained Semantic large classes is marked as M, and the mth Semantic large class is marked as Semantic_m，1≤m≤M；

A2, respectively acquiring the number of objects identified by each semantic in the semantic global map; among them, Semantic major class Semantic_mObject quantity tagging identified within the semantically global map

Step a3, randomly selecting a first preset number of clustering barycentric coordinates aiming at any semantic major category; wherein, aiming at any Semantic large class Semantic_mThe first preset number is

Step a4, calculating the distance between all grid coordinates marked with any semantic major category and each clustering gravity center coordinate;

step a5, dividing each grid coordinate into clusters formed by cluster gravity centers closest to the grid coordinate to obtain a first preset number of cluster combinations;

step a6, calculating the average value of all coordinates in all cluster combinations, and taking the obtained average value as the gravity center of the cluster combination corresponding to any semantic major category;

step a7, randomly selecting the first preset number of clustering barycentric coordinates again for any semantic major category, executing the step a 4-the step a6, and obtaining the barycentric of the clustering combination corresponding to any semantic major category again;

step a8, when the distance between the gravity center obtained in the step a6 and the gravity center obtained in the step a7 is smaller than a preset error, the step a9 is executed; otherwise, the step a7 is executed, and when the distance between the gravity center obtained in the step a6 and the gravity center obtained again is smaller than the preset error, the step a9 is executed;

step a9, calculating the distance between the barycentric coordinates of two clusters:

when the distance between any two clustering barycentric coordinates is smaller than a preset clustering distance parameter value, taking the clustering combination where the any two clustering barycentric coordinates are located as the same clustering combination, combining all coordinate data in the same clustering combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same clustering combination, obtaining an updated clustering barycentric coordinate combination, and turning to the step 10; otherwise, go to step a 9;

step a10, calculating the distance between the barycentric coordinates of all clusters after updating again:

when the distance between any two clustering barycentric coordinates is larger than a preset clustering distance parameter value, turning to a step a 11; otherwise, taking the cluster combination where any two cluster barycentric coordinates are located as the same cluster combination, combining all coordinate data in the same cluster combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same cluster combination, obtaining an updated cluster barycentric coordinate combination, and turning to step a 9;

step a11, the sort aggregation process is stopped.

Technical Field

The invention relates to the field of sweeping robots, in particular to a semantic-based sweeping robot self-adaptive sweeping method.

Background

Sweeping robots have become automatic cleaning equipment for more and more families. When the sweeping robot starts to work, the sweeping robot can complete the sweeping work aiming at the indoor area under manual control or automatically according to a preset sweeping mode. For example, the existing sweeping robot generally divides the area to be swept into a kitchen, a living room and a bedroom according to the room type, i.e. the sweeping mode of the sweeping robot has a sweeping kitchen mode, a sweeping living room mode and a sweeping bedroom mode. After the user places the sweeping robot in the room with the corresponding type, the user can start the corresponding sweeping mode on the sweeping robot according to the type of the current room, so that the sweeping robot can automatically complete sweeping for different rooms.

The chinese patent application CN110377014A discloses a general sweeping robot sweeping path planning method, which includes the following steps: (1) starting, and controlling the pose of the robot based on a closed loop; (2) searching for an object with laser reflection properties; (3) extracting and partitioning regions between the laser-reflective substances; (4) cleaning each subarea in sequence; (5) checking whether all the partitions are cleaned or not, if so, entering the next step, and if not, returning to the step (2); (6) and returning to the charging point. According to the sweeping path planning method of the sweeping robot in the patent application scheme, a mode of collecting environmental information and executing a sweeping task is adopted, so that time and battery energy consumption are saved, and the method is more humanized; in addition, the robot cleaning method performs subarea cleaning by combining with the structured geometric information of the target area, so that the cleaning direction of the robot is determined according to the geometric characteristics of each subarea, the energy consumption of the robot is reduced, the system is more robust, the cleaning robot can walk along the surface of a reflecting object when encountering the laser reflecting object, all areas near the laser reflecting object can be cleaned, and the target area is ensured to be completely cleaned. The invention patent application CN110377014A is to make the sweeping robot sweep the area near all laser reflection objects by searching for the objects with laser reflection properties and then partitioning according to the areas between the extracted laser reflection substances.

However, the invention patent application CN110377014A has some problems in the actual cleaning process: the sweeping robot always cleans the areas among the laser reflecting substances according to the existing fixed cleaning mode in the household cleaning process, and the actual space scene of the areas among the laser reflecting substances is not considered. If the sweeping robot utilizes the preset same sweeping working mode, and the zones between different laser reflecting substances in different room scenes in a house are swept according to the zone sweeping mode based on the zones between the laser reflecting substances, for example, the zones between objects such as a dining table, a chair, a carpet, a closestool and the like are swept, the sweeping robot cannot adaptively divide the zones between the objects in the sweeping process, cannot be adjusted to the corresponding sweeping mode, and finally cannot achieve a good cleaning effect.

Disclosure of Invention

The invention aims to solve the technical problem of providing a self-adaptive sweeping method of a sweeping robot based on semantics in the prior art.

The technical scheme adopted by the invention for solving the technical problems is as follows: a self-adaptive sweeping method of a sweeping robot based on semantics is characterized by comprising the following steps:

step 1, acquiring a global map of a scene area to be cleaned; wherein the global map is a rasterized map;

step 2, identifying and obtaining different object types in the global map;

step 3, respectively identifying various object types in the global map by using different semantics to obtain a semantization global map subjected to semantization identification processing;

step 4, corresponding numerical values are given to different semantics, the numerical values corresponding to the semantics are respectively filled into grids corresponding to the object types in the semantic global map, and the numerical global map after numerical value filling processing is obtained;

step 5, performing classification and aggregation processing on the regions with the same semantics in the numerical global map respectively to obtain a classified and aggregated global map; the regions with the same semantics in the aggregated global map are regions of the same type;

step 6, carrying out block division on grids in the same type of regions in the aggregated global map according to grid outlines to form a global map with different semantic cleaning regions, and taking the global map with different semantic cleaning regions as a target global map;

and 7, respectively executing corresponding cleaning working modes by the cleaning robot aiming at different semantic cleaning areas in the target global map.

According to the self-adaptive sweeping robot sweeping method based on the semantics, the global map of a scene to be swept is obtained, different object types in the global map are identified by different voices, then numerical values corresponding to different semantics are filled into grids corresponding to the object types, the grids with the same semantics are classified and aggregated by a classification and aggregation method, and the regions after classification and aggregation are divided according to the grid outline blocks, so that the sweeping robot can automatically execute corresponding sweeping working modes aiming at different semantic sweeping regions in a target global map, the sweeping robot can automatically recognize and adjust the sweeping working modes aiming at different sweeping regions, and the intelligent automatic sweeping efficiency of the sweeping robot is improved.

In order to meet the actual needs of the areas to be cleaned in different cleaning scenes, in the semantic-based sweeping robot adaptive cleaning method, in step 7, the sweeping robot executes different cleaning work modes for different semantic cleaning areas in the target global map.

For the to-be-cleaned scene of the sweeping robot in the invention, optionally, the to-be-cleaned scene area is at least one of a living room, a bedroom, a kitchen, a dining room and a toilet.

In order to finish the cleaning of a scene to be cleaned in the shortest time, in the semantic-based sweeping robot adaptive cleaning method, in step 7, an optimal cleaning path is respectively planned for different semantic cleaning areas in the target global map, and the sweeping robot cleans the corresponding semantic cleaning areas according to the optimal cleaning paths.

Further, in the semantic-based adaptive cleaning method for the cleaning robot, in the step 5, the aggregated global map is obtained in the following steps a 1-a 4:

a1, acquiring the total number of semantic large classes in the semantic global map; the total quantity of the obtained Semantic large classes is marked as M, and the mth Semantic large class is marked as Semantic_m，1≤m≤M；

A2, respectively acquiring the number of objects identified by each semantic in the semantic global map; among them, Semantic major class Semantic_mThe number of objects identified within the semantically global map is labeled N_Semanticm，N_Semanticm≥1；

Step a3, randomly selecting a first preset number of clustering barycentric coordinates aiming at any semantic major category; wherein, aiming at any Semantic large class Semantic_mThe first preset number is N_Semanticm；

Step a4, calculating the distance between all grid coordinates marked with any semantic major category and each clustering gravity center coordinate;

step a5, dividing each grid coordinate into clusters formed by cluster gravity centers closest to the grid coordinate to obtain a first preset number of cluster combinations;

step a6, calculating the average value of all coordinates in all cluster combinations, and taking the obtained average value as the gravity center of the cluster combination corresponding to any semantic major category;

step a7, randomly selecting the first preset number of clustering barycentric coordinates again for any semantic major category, executing the step a 4-the step a6, and obtaining the barycentric of the clustering combination corresponding to any semantic major category again;

step a8, when the distance between the gravity center obtained in the step a6 and the gravity center obtained in the step a7 is smaller than a preset error, the step a9 is executed; otherwise, the step a7 is executed, and when the distance between the gravity center obtained in the step a6 and the gravity center obtained again is smaller than the preset error, the step a9 is executed;

step a9, calculating the distance between the barycentric coordinates of two clusters:

when the distance between any two clustering barycentric coordinates is smaller than a preset clustering distance parameter value, taking the clustering combination where the any two clustering barycentric coordinates are located as the same clustering combination, combining all coordinate data in the same clustering combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same clustering combination, obtaining an updated clustering barycentric coordinate combination, and turning to the step 10; otherwise, go to step a 9;

step a10, calculating the distance between the barycentric coordinates of all clusters after updating again:

when the distance between any two clustering barycentric coordinates is larger than a preset clustering distance parameter value, turning to a step a 11; otherwise, taking the cluster combination where any two cluster barycentric coordinates are located as the same cluster combination, combining all coordinate data in the same cluster combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same cluster combination, obtaining an updated cluster barycentric coordinate combination, and turning to step a 9;

step a11, the sort aggregation process is stopped.

Compared with the prior art, the invention has the advantages that: according to the self-adaptive sweeping robot sweeping method based on the semantics, the global map of a scene to be swept is obtained, different object types in the global map are identified by different voices, then numerical values corresponding to different semantics are filled into grids corresponding to the object types, the grids with the same semantics are classified and aggregated by a classification and aggregation method, and the regions after classification and aggregation are divided according to the grid outline blocks, so that the sweeping robot can automatically execute corresponding sweeping work modes aiming at different semantic sweeping regions in a target global map, the sweeping robot can automatically recognize different sweeping regions and adjust the sweeping work modes, and the intelligent automatic sweeping efficiency of the sweeping robot is improved.

Drawings

Fig. 1 is a schematic flow chart of a semantic-based self-adaptive cleaning method for a sweeping robot in an embodiment of the invention.

Detailed Description

The invention is described in further detail below with reference to the accompanying examples.

As shown in fig. 1, the embodiment provides a semantic-based adaptive cleaning method for a cleaning robot. Here, the present embodiment is described with a living room and a restaurant in a home as a scene area to be cleaned. Specifically, the semantic-based sweeping robot self-adaptive sweeping method comprises the following steps:

step 1, acquiring a global map of a scene area to be cleaned; wherein, the global map is a rasterized map; the global map can be acquired by adopting sensing devices such as a laser radar or a depth/binocular camera; that is, a rasterized global map for the living room and the restaurants may be obtained in step 1;

step 2, identifying and obtaining different object types in the global map; here, tables, chairs, and carpets in the global map can be identified by the identification process for the acquired global map of the living room and the dining room;

step 3, respectively identifying various object types in the global map by using different semantics to obtain a semantization global map subjected to semantization identification processing;

and aiming at the dining table, the chair and the carpet identified in the global map, different semantics are respectively adopted for identification. For example, if the definition 1 is a table and chair object and the definition 2 is a ground attachment, the table and the chair identified in the global map are identified by the definition 1, and the carpet identified in the global map is identified by the definition 2;

step 4, corresponding numerical values are given to different semantics, the numerical values corresponding to the semantics are respectively filled into grids corresponding to the object types in the semantic global map, and the numerical global map after numerical value filling processing is obtained;

for example, after defining semantic 1, the embodiment further makes the assigned numerical value corresponding to semantic 1 be 1, and the assigned numerical value corresponding to semantic 2 be 2, and then fills the numerical value "1" into the grid occupied by the dining table in the global map, and simultaneously fills the grid occupied by the chair in the global map with the numerical value "1"; in addition, the value "2" is filled into the grid occupied by the carpet in the global map;

step 5, classifying and aggregating regions with the same semantics in the digitized global map in the step 4 respectively to obtain an aggregated global map after classification and aggregation; the regions with the same semantics in the aggregated global map are regions of the same type;

specifically, it can be known in this embodiment that the dining table area and the chair area in the numerical global map, where the numerical values are all "1", have the same semantic 1, and the carpet area in the numerical global map, where the numerical value is "2", has the semantic 2, so that the area with the numerical value of "1" belongs to the same cluster, and the area with the numerical value of "2" belongs to a single cluster, so as to obtain the aggregated global map in this embodiment;

step 6, carrying out block division on grids in the same type of regions in the aggregated global map according to grid outlines to form a global map with different semantic cleaning regions, and taking the global map with the different semantic cleaning regions as a target global map;

that is, after the first cluster with the numerical value of "1" is obtained by executing the step 5, the first cluster is divided along the contour of the dining table area grid and the chair area grid, so that an area to be cleaned, which comprises the dining table area grid with the semantic 1 and the chair area grid with the semantic 1, is formed; along the outline of the carpet area grid with the value of "2", another area B to be cleaned is formed;

and 7, respectively executing corresponding cleaning working modes by the cleaning robot aiming at different semantic cleaning areas in the target global map. After the two different areas a and B to be cleaned are obtained in step 6, the sweeping robot of this embodiment performs cleaning treatment on the area a to be cleaned and the area B to be cleaned respectively by using the preset cleaning operation mode.

Of course, here the sweeping robot in step 7 performs different sweeping modes of operation for different semantic sweeping areas within the target global map. For example, in this embodiment:

when the sweeping robot obtains that the area A to be swept is a restaurant according to the semantic-based processing mode, the sweeping robot starts a suction and mopping integrated sweeping working mode, and the executed optimal sweeping path adopts a Y-shaped reciprocating mode;

when the sweeping robot obtains the carpet of which the area B to be swept is the living room according to the semantic-based processing mode, the sweeping robot only starts a strong suction mode, and the executed optimal sweeping path adopts a roundabout Chinese character 'gong' mode.

In step 7, according to the cleaning requirement, respectively planning an optimal cleaning path for each different semantic cleaning area in the target global map, and respectively cleaning the corresponding semantic cleaning area by the cleaning robot according to each optimal cleaning path.

It should be noted that, in order to accurately perform classification and aggregation processing on the obtained digitized global map to obtain an aggregated global map for a scene area to be cleaned, the embodiment processes the obtained aggregated global map in the following steps a1 to a 4:

a1, acquiring the total number of semantic large classes in the semantic global map; the total quantity of the obtained Semantic large classes is marked as M, and the mth Semantic large class is marked as Semantic_m，1≤m≤M；

Specifically, in this embodiment, the semantic major categories of the scene area to be cleaned are respectively semantic 1 and semantic 2, that is, the first semantic major category is semantic 1, and the second semantic major category is semantic 2, so that the total number M of the acquired semantic major categories is 2;

step a2, respectively obtaining objects marked by each semantic in the semantic global mapThe number of bodies; among them, Semantic major class Semantic_mObject quantity tagging identified within a semantically-oriented global map

Because the object types identified by the semantic 1 are dining tables and chairs, and the object types identified by the semantic 2 are carpets, the number of objects identified by the semantic 1 in the semantic global map is 2, and the number of objects identified by the semantic 2 in the semantic global map is 1;

step a3, randomly selecting a first preset number of clustering barycentric coordinates aiming at any semantic major category; wherein, aiming at any Semantic large class Semantic_mThe first preset number isFor classification and clustering, the semantic 1 is used as an example here to randomly select two cluster barycentric coordinates, and the two cluster barycentrics selected at this time are respectively marked asAnd

step a4, calculating the distance between all grid coordinates marked with any semantic major category and each clustering gravity center coordinate;

assume that, in the scene area to be cleaned in this embodiment, the total number of grids occupied by the dining table in the global map is I, and the ith grid coordinate mark occupied by the dining table is denoted by I1≤i≤I；

Assuming that the total number of grids occupied by the chair in the global map is J, the jth grid coordinate mark occupied by the chair is1≤j≤J；

Respectively calculating grid coordinates of the dining table according to the grid coordinates of the dining table and the grids of the chair, which are marked with the semantic 1With the above two cluster barycenters (cluster barycenter G)₁And G₂) Distance between coordinates and calculating each grid coordinate of the chairWith the above two cluster barycenters (cluster barycenter G)₁And G₂) The distance between the coordinates;

supposing grid coordinates of dining tableAnd cluster center of gravity G₁Distance between coordinates is markedGrid coordinate of dining tableAnd cluster center of gravity G₂Distance between coordinates is marked

Suppose chair grid coordinatesAnd cluster center of gravity G₁Distance between coordinates is markedGrid coordinate of dining tableAnd cluster center of gravity G₂Distance between coordinates is marked

Step a5, dividing each grid coordinate into clusters formed by cluster gravity centers closest to the grid coordinate to obtain a first preset number of cluster combinations;

for example, assume grid coordinatesThe cluster gravity center closest to the cluster is G₁Grid coordinateThe cluster gravity center closest to the cluster is G₂Grid coordinateThe cluster gravity center closest to the cluster is G₁Grid coordinateThe cluster gravity center closest to the cluster is G₂Then the grid coordinates are calculatedAnd grid coordinatesDivision into clustering centers of gravity G₁In the formed cluster, grid coordinates are setAnd grid coordinatesDivision into clustering centers of gravity G₂In the formed cluster; for other grid coordinates of the dining table and the chair, the rest can be done by analogy, and the description is omitted here;

step a6, calculating the average value of all coordinates in all cluster combinations, and taking the obtained average value as the gravity center of the cluster combination corresponding to any semantic major category;

specifically, in this embodiment, the calculation is located at the cluster gravity center G₁Formed cluster combination and located at cluster gravity center G₂Average value of all coordinates in the formed cluster combination, and then the obtained average value is used as the gravity center of the cluster combination corresponding to the semantic 1

Step a7, randomly selecting the gravity center coordinates of the clusters with the first preset number again for any semantic major category, and executing the step a 4-the step a6 to obtain the gravity center of the cluster combination corresponding to any semantic major category again;

after the step a6 is finished, two clustering barycentric coordinates are randomly selected againAndand obtaining the gravity center of the cluster combination corresponding to the semantic 1 again according to the mode of the steps a 4-a 6

Step a8, when the distance between the gravity center obtained in the step a6 and the gravity center obtained in the step a7 is smaller than a preset error, the step a9 is executed; otherwise, the step a7 is executed, and when the distance between the gravity center obtained in the step a6 and the gravity center obtained again is smaller than the preset error, the step a9 is executed;

in particular, if the center of gravityAnd center of gravityWhen the distance between the two clusters is smaller than the preset error, the gravity center coordinate obtained after clustering can be better used as the gravity center of the current cluster, and the initial clustering is performedWhen the class work is finished, the step a9 is carried out; otherwise, the step a7 is shifted again until the center of gravity obtained in the step a6When the distance between the gravity center and the obtained gravity center is smaller than the preset error, the step a9 is carried out;

step a9, calculating the distance between the barycentric coordinates of two clusters:

when the distance between any two clustering barycentric coordinates is smaller than a preset clustering distance parameter value, taking the clustering combination where the any two clustering barycentric coordinates are located as the same clustering combination, combining all coordinate data in the same clustering combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same clustering combination, obtaining an updated clustering barycentric coordinate combination, and turning to the step 10; otherwise, go to step a 9; when the distance between any two clustering barycentric coordinates is smaller than a preset clustering distance parameter value, the two clustering distances are closer, and the two clusters can be combined into one class for processing, so that the total number of clusters is reduced;

step a10, calculating the distance between the barycentric coordinates of all clusters after updating again:

when the distance between any two clustering barycentric coordinates is larger than a preset clustering distance parameter value, turning to a step a 11; otherwise, taking the cluster combination where any two cluster barycentric coordinates are located as the same cluster combination, combining all coordinate data in the same cluster combination, taking the average value of all the combined coordinate data as the barycentric coordinate of the same cluster combination, obtaining an updated cluster barycentric coordinate combination, and turning to step a 9;

step a11, the sort aggregation process is stopped.

Of course, the scene area to be cleaned in this embodiment may also be at least one of a living room, a bedroom, a kitchen, a dining room, and a bathroom.