Cleaning control method

Abstract

The present invention discloses a cleaning control method, which extracts foreground object features and scene features from a captured image; detects whether the foreground object is an obstacle based on the extracted foreground object features; if the detection result is that it is an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path that avoids the obstacle point is reset; if the detection result is that it is impossible to determine whether it is an obstacle, a first conditional probability that the foreground object is an obstacle is further determined based on the extracted scene features and foreground object features, and if the first conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path that avoids the obstacle point is reset. In the present invention, scene features and foreground object features are combined, and obstacles are detected based on the conditional probability principle. The obstacle detection is more comprehensive, and it is convenient for the intelligent sweeping robot to change the cleaning path in time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent robots, and more specifically, to a cleaning control method for an intelligent sweeping robot. Background Art

[0002] With the development of artificial intelligence, more and more smart terminals have begun to enter people's lives. For example, a smart sweeping robot can automatically complete the tasks of sweeping, vacuuming, mopping the floor in the room, etc., absorb the debris on the floor into its own garbage storage box, and realize the cleaning of the floor. In the prior art, the smart sweeping robot usually collects the front image information through the front camera, detects whether there is an obstacle according to the collected front image information, and automatically avoids the obstacle if an obstacle is detected. However, the obstacle detection in the prior art is to extract the foreground object features according to the collected image. If the extracted foreground object features match all the features of the obstacle stored in advance, the foreground object is confirmed to be an obstacle. If the extracted foreground object features are only partial features of the obstacle, it is impossible to confirm whether the foreground object is an obstacle. However, when the smart sweeping robot is actually running, due to the aging of the camera device or the parameter setting, the collected image may be blurred. Only partial foreground object features may be collected through the blurred image, and the partial foreground object features cannot be matched with the pre-stored obstacle image, that is, the obstacle cannot be detected, resulting in the inability of the smart sweeping robot to change the cleaning path in time. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a cleaning control method for an intelligent sweeping robot, which has more comprehensive obstacle detection and facilitates the intelligent sweeping robot to change the cleaning path in time.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A cleaning control method, used for cleaning control of an intelligent cleaning robot, comprising:

[0006] Setting a first cleaning path for the intelligent sweeping robot to walk according to a target area to be cleaned by the intelligent sweeping robot;

[0007] Controlling the intelligent sweeping robot to clean according to the first cleaning path;

[0008] Collect images of the intelligent sweeping robot walking in front of it;

[0009] Extracting foreground object features and scene features from the acquired images;

[0010] Detecting whether the foreground object is an obstacle based on the extracted foreground object features;

[0011] If the detection result is an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path is reset to avoid the obstacle point;

[0012] If the detection result is that it is impossible to determine whether it is an obstacle, the first conditional probability that the foreground object is an obstacle is further determined based on the extracted scene features and foreground object features. If the first conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and the second cleaning path that avoids the obstacle point is reset.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] In the cleaning control method of the present invention, foreground object features and scene features are extracted from the captured image; whether the foreground object is an obstacle is detected based on the extracted foreground object features; if the detection result is that it is an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path that avoids the obstacle point is reset; and if the detection result is that it is impossible to determine whether it is an obstacle, for example, if the captured image is blurred, only part of the foreground object features may be captured through the blurred image, and it is impossible to determine whether the foreground object is an obstacle, then the first conditional probability that the foreground object is an obstacle is further determined based on the extracted scene features and foreground object features. If the first conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path that avoids the obstacle point is reset. In the present invention, scene features and foreground object features are combined, and obstacles are detected based on the conditional probability principle. The obstacle detection is more comprehensive, which facilitates the intelligent sweeping robot to change the cleaning path in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a specific embodiment of a cleaning control method of the present invention;

[0016] Figure 2 It is a schematic diagram of a specific embodiment of setting a cleaning path in a cleaning control method of the present invention;

[0017] Figure 3 It is a schematic diagram of a specific embodiment of encoding grid units in a cleaning control method of the present invention. DETAILED DESCRIPTION

[0018] refer to Figure 1 , which is a flow chart of a specific embodiment of a cleaning control method of the present invention. The method of this embodiment mainly includes the following steps:

[0019] Step S101, according to the target area to be cleaned by the intelligent sweeping robot, a first cleaning path for the intelligent sweeping robot to walk is set. In specific implementation, the path setting is a full coverage path setting within the target area. A variety of algorithms can be used to set the cleaning path, for example, a random coverage method, a Dijkstra algorithm, a neural network algorithm, etc. Figure 2 , this figure is a schematic diagram of a cleaning path implementation example set by a random covering method, and other methods can also be used for setting, which is not specifically limited here;

[0020] Step S102, controlling the intelligent sweeping robot to sweep along the first sweeping path, in specific implementation, Figure 2 Taking the set cleaning path as an example, the intelligent sweeping robot travels in one direction until turning as one cleaning process, and it takes 4 cleaning processes to clean one circle, which will not be repeated here;

[0021] Step S103, collecting images in front of the intelligent sweeping robot. In specific implementation, it is necessary to set an image acquisition device in front of the intelligent sweeping robot body. The image acquisition device can be a camera, a camera, etc. In addition, the image acquisition device can collect images in front of the intelligent sweeping robot in real time, and can also determine the large object as the object to be detected when the intelligent sweeping robot cleans a large object. There is no specific limitation here.

[0022] Step S104, extracting foreground object features and scene features from the collected image. In specific implementation, various methods can be used to extract features from the collected image. For example, the foreground object features can be extracted by converting the image into a binary image, thereby dividing the image into a foreground part and a background part, and then the binary image is superimposed with the original image to obtain a foreground image, and then the foreground object features can be extracted based on the foreground image. The method for extracting the foreground object features is not specifically limited here. In addition, the scene features can also be extracted according to the above method, which will not be repeated here.

[0023] Step S105, detecting whether the foreground object is an obstacle according to the extracted foreground object features. In specific implementation, detecting whether the foreground object is an obstacle according to the foreground object features may be performed by a feature point matching method, that is, the obstacle features are determined in advance, and the determined foreground object features are matched with the obstacle features. If the two match, it is determined that the foreground object is an obstacle. If the two do not match, it is determined that the foreground object is not an obstacle.

[0024] In this embodiment, in step S106, if the captured image is clear and the detection result shows that the foreground object is an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path is reset to avoid the obstacle point. If the detection result shows that the foreground object is not an obstacle, cleaning is continued according to the first cleaning path.

[0025] In addition, for example, if the captured image is blurred, the extracted foreground object features are only part of all the features, and it is impossible to determine whether the foreground object is an obstacle based on the extracted foreground object features. For this reason, in step S107 of this embodiment, when the detection result is that it is impossible to determine whether the foreground object is an obstacle, the first conditional probability that the foreground object is an obstacle is determined based on the extracted scene features and foreground object features. If the first conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and the second cleaning path that avoids the obstacle point is reset; if the first conditional probability is less than the preset threshold, it can be determined that the foreground object is not an obstacle, and cleaning continues according to the first cleaning path.

[0026] The following is a detailed description of the method of detecting obstacles based on conditional probability in the present invention. The principle of detecting obstacles based on conditional probability in the present invention is to perform detection by taking scene features and foreground object features as detection constraints. Specifically, in this embodiment, the first conditional probability that the foreground object is an obstacle is further determined based on the extracted scene features and foreground object features, specifically in the following manner, namely:

[0027] Combine the scene features and foreground object features into various conditions in advance, determine the conditional probability of the foreground object being an obstacle under each condition and save it;

[0028] Determine corresponding conditions according to the extracted scene features and foreground object features;

[0029] According to the determined condition, the pre-stored conditional probability information is queried to obtain a first conditional probability corresponding to the condition.

[0030] Let's take a simple example to illustrate. For example, assuming that the intelligent sweeping robot is in an environment where there are two scene features, A1 and A2, and two foreground features, B1 and B2. The scene features are combined with the foreground features to obtain four conditions, namely: A1B1, A1B2, A2B1 and A2B2. The threshold is set to 80%. By training and testing the samples, it is predetermined that the probability of the foreground object being an obstacle under the A1B1 condition is 40%, the probability of the foreground object being an obstacle under the A1B2 condition is 90%, the probability of the foreground object being an obstacle under the A2B1 condition is 75%, and the probability of the foreground object being an obstacle under the A2B2 condition is 60%. In the prior art, only the condition matching B1 is used. Only when both foreground object features A1 and B2 are present can the foreground object be determined to be an obstacle. When only the foreground object feature B2 is extracted, it is impossible to determine whether the foreground object is an obstacle or to directly determine that the foreground object is not an obstacle. In the present invention, the extracted foreground object feature B2 is combined with the extracted scene feature A1 to determine the corresponding condition to be A1B2. The pre-saved conditional probability information is further queried to determine that the conditional probability that the foreground object corresponding to the condition A1B2 is an obstacle is 90%, which is greater than the pre-set threshold of 80%. The foreground object can be determined to be an obstacle. In the present invention, by combining the scene feature with the foreground object feature as the detection constraint condition, the obstacle detection can be made more comprehensive, which is convenient for the intelligent sweeping robot to change the cleaning path in time.

[0031] It should be noted that, as another preferred embodiment, the present invention also extracts reference object features from the collected image;

[0032] If the detection result is that it is impossible to determine whether it is an obstacle, the second conditional probability that the foreground object is an obstacle is further determined based on the extracted scene features, reference object features and foreground object features. If the second conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and the third cleaning path that avoids the obstacle point is reset.

[0033] It should be noted that the second conditional probability of further determining that the foreground object is an obstacle based on the extracted scene features, reference object features and foreground object features can also be determined in the following manner, namely:

[0034] Combine the scene features, reference object features and foreground object features into various conditions in advance, determine the conditional probability of the foreground object being an obstacle under each condition and save it;

[0035] Determine corresponding conditions according to the extracted scene features, reference object features and foreground object features;

[0036] According to the determined condition, the pre-stored conditional probability information is queried to obtain a second conditional probability corresponding to the condition.

[0037] In addition, in order to improve the working efficiency of the intelligent sweeping robot, as a preferred embodiment, the present invention also includes:

[0038] The target area to be cleaned by the intelligent sweeping robot is divided into grid units, where the grid units are divided into free grid units and obstacle grid units. The free grid units are areas where there are free passages, and the obstacle grid units are areas where there are obstacles. Figure 3 In this embodiment, the grid cells can be encoded, the free grid cells are encoded as 1, and the obstacle grid cells are encoded as 0. The encoding allows the intelligent sweeping robot to quickly identify the grid cells to reduce the cleaning time;

[0039] It should be noted that in this embodiment, the smart sweeping robot is controlled to clean in the fast cleaning mode in the grid free unit, and is controlled to clean in the fine cleaning mode in the obstacle grid unit. In this way, on the one hand, the fast cleaning mode is adopted in the grid free unit to ensure the working efficiency of the smart sweeping robot. At the same time, there are generally more kinds of garbage in the obstacles. Therefore, the fine cleaning mode is adopted in the obstacle grid unit to sweep more thoroughly.

[0040] In addition, in the present invention, after cleaning is completed, the grid unit coding information of the target area is saved, and the cleaning environment map is updated according to the saved grid unit coding information of multiple cleanings. When the next cleaning is performed, the cleaning path is set according to the updated environment map, which will not be repeated here.

Claims

1. A cleaning control method for controlling the cleaning of an intelligent cleaning robot, characterized in that: include: Setting a first cleaning path for the intelligent sweeping robot to walk according to a target area to be cleaned by the intelligent sweeping robot; Controlling the intelligent sweeping robot to clean according to the first cleaning path; Collect images of the intelligent sweeping robot walking in front of you; Extracting foreground object features and scene features from the acquired images; Detecting whether the foreground object is an obstacle based on the extracted foreground object features; When the detection result is an obstacle, the area where the foreground object is located is marked as an obstacle point, and a second cleaning path is reset to avoid the obstacle point; When the detection result is that it is impossible to determine whether it is an obstacle, the first conditional probability that the foreground object is an obstacle is determined based on the extracted scene features and foreground object features. If the first conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and the second cleaning path is reset to avoid the obstacle point; The first conditional probability of determining that the foreground object is an obstacle according to the extracted scene features and foreground object features specifically includes: Combine the scene features and foreground object features into various conditions in advance, determine the conditional probability of the foreground object being an obstacle under each condition and save it; Determine corresponding conditions according to the extracted scene features and foreground object features; According to the determined condition, the pre-stored conditional probability information is queried to obtain a first conditional probability corresponding to the condition.

2. The method according to claim 1, characterized in that It also includes extracting reference object features from the acquired image; If the detection result is that it is impossible to determine whether it is an obstacle, the second conditional probability that the foreground object is an obstacle is determined based on the extracted scene features, reference object features and foreground object features. If the second conditional probability is greater than a preset threshold, the foreground object is determined to be an obstacle, the area where the foreground object is located is marked as an obstacle point, and the third cleaning path is reset to avoid the obstacle point.

3. The method according to claim 1, characterized in that Also includes: The target area to be cleaned by the intelligent sweeping robot is divided into grid units, where the grid units are divided into free grid units and obstacle grid units. The free grid units are areas for free passage, and the obstacle grid units are areas where obstacles exist.

4. The method according to claim 3, characterized in that: In the free grid unit, the intelligent sweeping robot is controlled to clean in a fast sweeping mode, and in the obstacle grid unit, the intelligent sweeping robot is controlled to clean in a precise sweeping mode.

Images