Robot path planning method based on machine vision

Through machine vision and deep learning technology, paths are detected and adjusted in real time, the problem of insufficient adaptability of traditional path planning in dynamic environments is solved, and more flexible and intelligent robot navigation is achieved.

CN118502452BActive Publication Date: 2024-11-12WUXI TUCHUANG INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410964957.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-12
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

Traditional robot path planning methods assume that the environment is static and cannot effectively adapt to complex, unknown or dynamically changing environments, resulting in insufficient adaptability and intelligence in actual applications.

Method used

Using a machine vision-based method, the route field of view images are collected through the robot's camera, and the obstacle target recognizer of deep learning is used to detect obstacles in real time, adjust the travel angle and re-detection. If there are no obstacles, move along a new angle to achieve dynamic Path planning.

Benefits of technology

The flexibility and intelligence of robot path planning are improved, so that it can effectively avoid obstacles and reach the target position to adapt to changes in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118502452B_ABST
    Figure CN118502452B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a robot path planning method based on machine vision. The method includes: acquiring a first travel route field of view image collected by a camera of the robot, and then using a deep learning-based obstacle target identifier to perform obstacle target identification to obtain a first identification result; adjusting the travel angle of the robot, and again collecting a second travel route field of view image through the camera; using the deep learning-based obstacle target identifier to perform obstacle target identification to obtain a second identification result; and, in response to the second identification result that there is no obstacle object, driving the robot to move forward along the adjusted travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to more effectively avoid obstacles and reach the target position, thereby improving the flexibility and intelligence of the robot's path planning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine vision, and in particular to a robot path planning method based on machine vision. Background Art

[0002] Robot path planning refers to the process of determining the optimal path from the starting point to the target point of the robot through an algorithm in a given environment. Robot path planning is of great significance for applications such as autonomous navigation robots and self-driving vehicles.

[0003] However, traditional robot path planning methods are mainly based on geometry and planning algorithms, and usually assume that the environment is static, that is, obstacles do not move or change. This assumption is often not true in actual scenarios, because obstacles in real environments may move, appear, and disappear, and real-time adaptation to environmental changes is required. In other words, traditional planning methods often rely on static map information for path planning, and have poor adaptability to complex, unknown, or dynamically changing environments, and cannot achieve intelligent and flexible path planning.

[0004] Therefore, a robot path planning solution based on machine vision is desired. Summary of the invention

[0005] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] In a first aspect, the present disclosure provides a robot path planning method based on machine vision, the method comprising:

[0007] Acquire a first travel route field of view image captured by a camera of the robot;

[0008] Using a deep learning-based obstacle target identifier to perform obstacle target identification on the first traveling route field image to obtain a first identification result, wherein the first identification result is used to indicate whether an obstacle object exists;

[0009] Adjusting the travel angle of the robot, and collecting a second travel route field image through the camera again; and

[0010] Using the deep learning-based obstacle target identifier to perform obstacle target identification on the second traveling route field image to obtain a second identification result; and

[0011] In response to the second recognition result that there is no obstacle object, the robot is driven to move forward along the adjusted travel angle.

[0012] Optionally, an obstacle target identifier based on deep learning is used to perform obstacle target identification on the first route field of view image to obtain a first identification result, and the first identification result is used to indicate whether there is an obstacle object, including: performing brightness compensation on the first route field of view image to obtain a first route field of view image after brightness compensation; performing feature extraction on the first route field of view image after brightness compensation through a deep neural network to obtain a route field of view shallow feature map and a route field of view semantic feature map; passing the route field of view shallow feature map and the route field of view semantic feature map through a speech-guided attention feature selection network to obtain a semantic-guided route field of view multi-scale feature map; passing the semantic-guided route field of view multi-scale feature map through a regional transfer module based on an attention mechanism to obtain a foreground-highlighted semantic-guided route field of view multi-scale feature map; and determining the first identification result based on the foreground-highlighted semantic-guided route field of view multi-scale feature map.

[0013] Optionally, the deep neural network is a progressive feature pyramid network.

[0014] Optionally, the shallow feature map of the route vision and the semantic feature map of the route vision are passed through a speech-guided attention feature selection network to obtain a semantic-guided multi-scale feature map of the route vision, including: performing global mean pooling processing on each feature matrix along the channel dimension in the semantic feature map of the route vision to obtain a semantic pooling feature vector of the route vision; passing the semantic pooling feature vector of the route vision through a fully connected module to perform full connection encoding on the semantic pooling feature vector of the route vision to obtain a semantic fully connected feature vector of the route vision; performing weighted fusion based on the channel dimension on the semantic fully connected feature vector of the route vision and the shallow feature map of the route vision to obtain a semantic multi-level fusion feature map of the route vision; and combining the multi-level fusion feature map of the route vision with the shallow feature map of the route vision. The vision semantic feature map is fused to obtain the semantically guided route vision multi-scale feature map; wherein, each feature matrix along the channel dimension in the route vision semantic feature map is globally mean pooled to obtain the route vision semantic pooling feature vector, including: calculating the global mean of each feature matrix along the channel dimension in the route vision shallow feature map to obtain the route vision semantic pooling feature vector; wherein, the route vision semantic fully connected feature vector and the route vision shallow feature map are weighted fused based on the channel dimension to obtain the route vision semantic multi-level fusion feature map, including: using the route vision semantic fully connected feature vector to perform weighted processing on the route vision shallow feature map along the channel dimension to obtain the route vision semantic multi-level fusion feature map.

[0015] Optionally, the route view semantic pooling feature vector is passed through a fully connected module to perform fully connected encoding on the route view semantic pooling feature vector to obtain a route view semantic fully connected feature vector, including: after matrix multiplication of the route view semantic pooling feature vector and the predetermined weight coefficient matrix, the obtained feature vector is added to the predetermined offset vector by position to obtain the modulated route view semantic pooling feature vector; using The activation function activates the modulated route view semantic pooling feature vector to obtain the route view semantic fully connected feature vector.

[0016] Optionally, the semantically guided route field of view multi-scale feature map is passed through a regional transfer module based on an attention mechanism to obtain a foreground-highlighted semantically guided route field of view multi-scale feature map, including: calculating a semantically guided route field of view spatial attention feature matrix of the semantically guided route field of view multi-scale feature map; masking the semantically guided route field of view spatial attention feature matrix based on a predetermined threshold to obtain a masked semantically guided route field of view spatial attention feature matrix; calculating the positional point multiplication between each feature matrix along the channel dimension in the semantically guided route field of view multi-scale feature map and the masked semantically guided route field of view spatial attention feature matrix to obtain the foreground-highlighted semantically guided route field of view multi-scale feature map.

[0017] Optionally, calculating the semantically guided route field of view spatial attention feature matrix of the semantically guided route field of view multi-scale feature map includes: performing global mean pooling processing on the semantically guided route field of view multi-scale feature map along the channel dimension to obtain the semantically guided route field of view global mean pooling feature matrix; inputting the semantically guided route field of view global mean pooling feature matrix into a Sigmoid activation unit to obtain the semantically guided route field of view spatial attention feature matrix.

[0018] Optionally, the semantically guided route field of view spatial attention feature matrix is ​​masked based on a predetermined threshold to obtain a masked semantically guided route field of view spatial attention feature matrix, including: in response to each position feature value in the semantically guided route field of view spatial attention feature matrix being greater than or equal to the predetermined threshold, taking the original value of the position feature value, otherwise setting it to zero.

[0019] Optionally, determining the first recognition result based on the foreground-prominent semantics-guided route field of view multi-scale feature map includes: passing the foreground-prominent semantics-guided route field of view multi-scale feature map through a classifier to obtain the first recognition result.

[0020] The above technical solution is adopted to obtain the first route field image collected by the robot's camera, and then use the deep learning-based obstacle target identifier to perform obstacle target identification to obtain a first identification result; adjust the robot's travel angle, and collect the second route field image through the camera again; use the deep learning-based obstacle target identifier to perform obstacle target identification to obtain a second identification result; and, in response to the second identification result that there is no obstacle object, drive the robot to move forward along the adjusted travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to more effectively avoid obstacles and reach the target position, thereby improving the flexibility and intelligence of the robot's path planning.

[0021] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0023] Figure 1 The present invention is a flowchart of a robot path planning method based on machine vision according to an exemplary embodiment.

[0024] Figure 2 The invention is a block diagram of a robot path planning system based on machine vision according to an exemplary embodiment.

[0025] Figure 3 It is a block diagram of an electronic device according to an exemplary embodiment.

[0026] Figure 4 This is an application scenario diagram of a robot path planning method based on machine vision according to an exemplary embodiment. DETAILED DESCRIPTION

[0027] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0029] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0030] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0031] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0032] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0033] Robot Path Planning is a very important field that involves multiple disciplines, including but not limited to robotics, artificial intelligence, computer science, and mathematics. In robot path planning, the main goal is to find a path from the starting point to the end point. This path needs to meet the following conditions:

[0034] 1. The path must be something the robot can realistically walk, given the robot’s physical size and locomotion capabilities.

[0035] 2. The path should avoid obstacles and potential danger areas.

[0036] 3. Under the premise of satisfying the first two conditions, the path should be as short as possible to reduce time and energy consumption, which usually involves minimizing the cost function.

[0037] 4. For some applications, such as autonomous vehicles, path planning needs to be done quickly in a real-time environment.

[0038] 5. The path planning algorithm should be able to adapt to environmental changes, such as dynamically appearing obstacles.

[0039] Robot path planning algorithms can be divided into several main categories: Graph-based planning: using a search algorithm to find a path on an abstract representation of the environment (usually a graph). Typical algorithms include Dijkstra's algorithm, A* search algorithm, etc.; Sampling-based planning: does not rely on an explicit representation of the environment, but explores the environment through random sampling. Representative algorithms include Rapid Exploration Random Trees (RRT) and Artificial Potential Fields; Vector Field Histogram (VFH): combines the characteristics of graph-based planning and sampling-based planning, and guides the movement of the robot by constructing a vector field in a local area of ​​the environment; Dynamic Programming: suitable for smaller-scale problems, and finds the optimal solution through recursive calculations.

[0040] The real-world environment is often dynamically changing, which requires the path planning system to be able to handle not only static obstacles, but also moving obstacles and changing environmental conditions. In a dynamic environment, the path planning algorithm needs to be able to respond quickly to update the path in real time to ensure that the robot can navigate safely and effectively. In addition to real-time performance, the path planning system also needs to have the ability to make intelligent decisions and be able to select the best or most feasible path among multiple possible paths. Planning methods that rely on static map information may not provide effective solutions when faced with environmental changes because they cannot predict or adapt to unknown or changing situations.

[0041] In order to achieve effective path planning in complex, unknown or dynamically changing environments, it is necessary to develop algorithms that can adapt to these conditions and handle the uncertainty and variability of the environment. Solving these problems usually requires combining technologies and methods from multiple fields such as robotics, artificial intelligence, machine learning, sensor fusion and control theory. With the development of technology, new path planning algorithms are constantly being proposed, which can better handle dynamic environments and provide more flexible and intelligent navigation solutions.

[0042] In order to solve the above problems, the present disclosure provides a robot path planning method based on machine vision, which obtains a first travel route field image collected by the robot's camera, and then uses a deep learning-based obstacle target identifier to perform obstacle target identification to obtain a first identification result; adjusts the robot's travel angle, and again collects a second travel route field image through the camera; uses the deep learning-based obstacle target identifier to perform obstacle target identification to obtain a second identification result; and, in response to the second identification result that there is no obstacle object, drives the robot to move forward along the adjusted travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to more effectively avoid obstacles and reach the target position, thereby improving the flexibility and intelligence of the robot's path planning.

[0043] The specific embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0044] Figure 1 is a flow chart of a robot path planning method based on machine vision according to an exemplary embodiment. Figure 1 As shown, the method includes:

[0045] Step 101, obtaining a first traveling route field image captured by a camera of the robot;

[0046] Step 102: using a deep learning-based obstacle target identifier to perform obstacle target identification on the first traveling route field of view image to obtain a first identification result, where the first identification result is used to indicate whether an obstacle object exists;

[0047] Step 103: adjusting the travel angle of the robot, and collecting a second travel route field image through the camera again;

[0048] Step 104: using the deep learning-based obstacle target identifier to perform obstacle target identification on the second traveling route field image to obtain a second identification result; and

[0049] Step 105: In response to the second recognition result that there is no obstacle object, drive the robot to move forward along the adjusted travel angle.

[0050] In response to the above technical problems, the technical concept of the present application is to collect field of view images of different travel routes in real time through the robot's camera, and introduce machine vision and artificial intelligence technology at the back end to analyze the field of view image of the first travel route, so as to determine whether there is an obstacle object in the first direction. When an obstacle is detected, the travel angle is adjusted, and obstacle target detection in other directions is performed again. If no obstacle is detected, the robot is controlled to move forward along the travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to more effectively avoid obstacles and reach the target position, thereby improving the flexibility and intelligence of the robot's path planning.

[0051] Specifically, in the technical solution of the present application, a robot path planning solution based on machine vision is proposed, which includes the steps of: acquiring a first travel route field of view image captured by a camera of the robot; using a deep learning-based obstacle target identifier to perform obstacle target identification on the first travel route field of view image to obtain a first identification result, and the first identification result is used to indicate whether there is an obstacle object; if the first identification result indicates that there is an obstacle object, adjusting the travel angle of the robot, and again capturing a second travel route field of view image through the camera; if the first identification result indicates that there is no obstacle object, the subsequent step of adjusting the travel angle is not performed; using the deep learning-based obstacle target identifier to perform obstacle target identification on the second travel route field of view image to obtain a second identification result; in response to the second identification result that there is no obstacle object, driving the robot to move forward along the adjusted travel angle.

[0052] In one embodiment of the present disclosure, an obstacle target identifier based on deep learning is used to perform obstacle target identification on the first route field of view image to obtain a first identification result, and the first identification result is used to indicate whether there is an obstacle object, including: performing brightness compensation on the first route field of view image to obtain a brightness compensated first route field of view image; performing feature extraction on the brightness compensated first route field of view image through a deep neural network to obtain a route field of view shallow feature map and a route field of view semantic feature map; passing the route field of view shallow feature map and the route field of view semantic feature map through a speech-guided attention feature selection network to obtain a semantic-guided route field of view multi-scale feature map; passing the semantic-guided route field of view multi-scale feature map through a region transfer module based on an attention mechanism to obtain a foreground-highlighted semantic-guided route field of view multi-scale feature map; and determining the first identification result based on the foreground-highlighted semantic-guided route field of view multi-scale feature map.

[0053] Wherein, the deep neural network is a progressive feature pyramid network.

[0054] Furthermore, in the above-mentioned robot path planning scheme, in the process of using a deep learning-based obstacle target identifier to identify the first travel route field of view image to obtain the first identification result, in order to more accurately identify and detect obstacles in the direction of the first travel route, in the technical solution of the present application, first, the first travel route field of view image is brightness compensated to obtain the first travel route field of view image after brightness compensation. It should be understood that in robot path planning, image quality is crucial to the accuracy of obstacle target recognition. If the brightness of the image is uneven or there are too dark or too bright parts, it may affect the recognition effect of the obstacle, resulting in the robot being unable to accurately perceive the obstacles in the environment. Therefore, by performing brightness compensation on the image, the brightness distribution of the image can be made more uniform, highlighting the key information in the image, which is conducive to the subsequent obstacle target recognition work. In addition, the image after brightness compensation can improve the stability and robustness of the algorithm, help reduce errors caused by changes in illumination, and improve the performance and reliability of the robot path planning system.

[0055] Performing brightness compensation on the first traveling route field of view image according to the following brightness compensation formula to obtain the first traveling route field of view image after brightness compensation;

[0056] Wherein, the brightness compensation formula is:

[0057]

[0058] in, Convert the first traveling route visual field image from the RGB space to the pixel values ​​of each position in the V channel normalized image of the HSV space, A , B , C and D To tune the hyperparameters, It is the pixel value of each position in the first traveling route field image after the brightness compensation.

[0059] Next, considering that there is feature information of different levels and scales in the first direction of travel in the first travel route field of view image after brightness compensation, these features are important for obstacle recognition and detection in the first direction of travel. Therefore, in order to capture the multi-level and multi-scale feature information in the first travel route field of view image after brightness compensation, so as to more accurately identify the complex structures and targets in the image and provide a more comprehensive data basis for obstacle detection, in the technical solution of the present application, the first travel route field of view image after brightness compensation is further passed through the progressive feature pyramid network (AFPN) to obtain the shallow feature map of the travel route field of view and the semantic feature map of the travel route field of view. By applying the progressive feature pyramid network, shallow image features and deep semantic features in the first traveling direction can be extracted from the brightness-compensated first traveling route field of view image, wherein the shallow features can reflect the color, texture, edge and other features of different objects in the first traveling direction, which are of great significance for obstacle detection; and the deep semantic features can provide semantic information about object categories and different elements, which helps to more accurately identify key elements such as obstacles and roads in the environment, thereby better realizing the planning of the robot's travel path.

[0060] Then, considering that the shallow feature map of the route vision and the semantic feature map of the route vision respectively contain the shallow features and deep semantic features of different target objects in the first direction of travel, and the features of different levels and scales are of great significance for the recognition and detection of obstacles. Therefore, in order to integrate the feature information of different levels and scales in the field of view image in the first direction of travel, in the technical solution of the present application, the shallow feature map of the route vision and the semantic feature map of the route vision are further passed through the semantic guidance attention feature selection network to obtain the semantic guidance route vision multi-scale feature map. By using the semantic guidance attention feature selection network, the shallow features and deep semantic features in the first route can be effectively integrated. Specifically, it can guide the feature selection and expression process of the shallow feature map of the route vision according to the deep semantic feature information of the first route vision image, so that the multi-scale feature information related to the obstacle category and useful in the image can be better paid attention to, so as to more accurately understand and express the actual semantic content in the image, which helps to improve the expression ability and discrimination of the features, thereby improving the path planning system's ability to understand and recognize the environment.

[0061] The shallow feature map of the route vision and the semantic feature map of the route vision are processed by the speech-guided attention feature selection network using the following feature selection formula to obtain the semantic-guided route vision multi-scale feature map;

[0062] Among them, the feature selection formula is:

[0063]

[0064] in, is the first feature matrix of each feature matrix along the channel dimension in the shallow feature map of the route field of view The eigenvalues ​​of the positions, and are the height and width of the shallow feature map of the route field of view, respectively. is the feature value of each position in the semantic pooling feature vector of the route vision, is the semantic pooling feature vector of the route vision, represents the weight coefficient matrix, is the offset vector, for Activation function, is the semantic fully connected feature vector of the route vision, Indicates that the weighted operation is performed on the feature map using the weight vector. is the semantic feature map of the travel route field of view, is the shallow feature map of the vision of the travel route, It is a multi-scale feature map of the vision of the semantically guided route.

[0065] In one embodiment of the present disclosure, the shallow feature map of the route vision and the semantic feature map of the route vision are passed through a speech-guided attention feature selection network to obtain a semantic-guided multi-scale feature map of the route vision, including: performing global mean pooling processing on each feature matrix along the channel dimension in the semantic feature map of the route vision to obtain a semantic pooling feature vector of the route vision; passing the semantic pooling feature vector of the route vision through a fully connected module to perform full connection encoding on the semantic pooling feature vector of the route vision to obtain a semantic fully connected feature vector of the route vision; performing weighted fusion based on the channel dimension on the semantic fully connected feature vector of the route vision and the shallow feature map of the route vision to obtain a multi-level fusion feature map of the route vision semantics; and combining the multi-level fusion feature map of the route vision semantics and the shallow feature map of the route vision. The semantic feature map of the route vision is fused to obtain the semantically guided multi-scale feature map of the route vision; wherein, each feature matrix along the channel dimension in the semantic feature map of the route vision is globally mean pooled to obtain the semantic pooling feature vector of the route vision, including: calculating the global mean of each feature matrix along the channel dimension in the shallow feature map of the route vision to obtain the semantic pooling feature vector of the route vision; wherein, the semantic fully connected feature vector of the route vision and the shallow feature map of the route vision are weighted fused based on the channel dimension to obtain the semantic multi-level fusion feature map of the route vision, including: using the semantic fully connected feature vector of the route vision to perform weighted processing on the shallow feature map of the route vision along the channel dimension to obtain the semantic multi-level fusion feature map of the route vision.

[0066] Further, the route view semantic pooling feature vector is passed through a fully connected module to perform fully connected encoding on the route view semantic pooling feature vector to obtain a route view semantic fully connected feature vector, including: after matrix multiplication of the route view semantic pooling feature vector and the predetermined weight coefficient matrix, the obtained feature vector is added to the predetermined offset vector by position to obtain the modulated route view semantic pooling feature vector; using The activation function activates the modulated route view semantic pooling feature vector to obtain the route view semantic fully connected feature vector.

[0067] Furthermore, in order to further improve the ability to express the substantial semantics and features of obstacles in the image in the first direction of travel, so as to highlight the feature information of the foreground area of ​​the obstacles, so as to more accurately identify and detect the obstacles, in the technical solution of the present application, the semantically guided route field of view multi-scale feature map is further passed through a regional transfer module based on an attention mechanism to obtain a foreground highlighted semantically guided route field of view multi-scale feature map. It should be understood that through the processing of the regional transfer module based on the attention mechanism, the importance of different positions in the semantically guided route field of view multi-scale feature map in the obstacle identification task can be learned and captured, and the features of different areas can be weighted and adjusted, so as to focus more attention on important foreground areas, highlight foreground information, and enhance the semantic feature representation of important areas in the feature map. This helps the system better identify key targets and obstacles in the environment, improve the expressiveness and discrimination of features, and provide more accurate information for path planning.

[0068] In one embodiment of the present disclosure, the semantically guided route field of view multi-scale feature map is passed through a regional transfer module based on an attention mechanism to obtain a foreground-highlighted semantically guided route field of view multi-scale feature map, including: calculating a semantically guided route field of view spatial attention feature matrix of the semantically guided route field of view multi-scale feature map; masking the semantically guided route field of view spatial attention feature matrix based on a predetermined threshold to obtain a masked semantically guided route field of view spatial attention feature matrix; calculating the positional point multiplication between each feature matrix along the channel dimension in the semantically guided route field of view multi-scale feature map and the masked semantically guided route field of view spatial attention feature matrix to obtain the foreground-highlighted semantically guided route field of view multi-scale feature map.

[0069] Furthermore, the semantically guided route field of view spatial attention feature matrix of the semantically guided route field of view multi-scale feature map is calculated, including: performing global mean pooling processing on the semantically guided route field of view multi-scale feature map along the channel dimension to obtain the semantically guided route field of view global mean pooling feature matrix; inputting the semantically guided route field of view global mean pooling feature matrix into a Sigmoid activation unit to obtain the semantically guided route field of view spatial attention feature matrix.

[0070] Furthermore, the semantically guided route field of view spatial attention feature matrix is ​​masked based on a predetermined threshold to obtain a masked semantically guided route field of view spatial attention feature matrix, including: in response to each position feature value in the semantically guided route field of view spatial attention feature matrix being greater than or equal to the predetermined threshold, taking the original value of the position feature value, otherwise setting it to zero.

[0071] Then, the multi-scale feature map of the vision of the route guided by the foreground-highlighted semantics is passed through a classifier to obtain the first recognition result. That is to say, the multi-scale feature representation information of the vision of the route guided by the image semantics after foreground highlighting is used for classification processing to identify obstacles on the first route, thereby determining whether there are obstacle objects. In this way, when an obstacle is detected, the travel angle is adjusted, and obstacle target detection in other directions is performed again. If no obstacle is detected, the robot is controlled to move forward along the travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to more effectively avoid obstacles and reach the target position, thereby improving the flexibility and intelligence of the robot's path planning.

[0072] In one embodiment of the present disclosure, determining the first recognition result based on the foreground-prominent semantically-guided route field of view multi-scale feature map includes: passing the foreground-prominent semantically-guided route field of view multi-scale feature map through a classifier to obtain the first recognition result.

[0073] In a preferred example of the above technical solution, the multi-scale feature map of the foreground highlighting semantic guidance route field of view is passed through a classifier to obtain the first recognition result, specifically including:

[0074] The foreground highlighting semantics guiding the route vision multi-scale feature map is clustered and optimized using the following formula to obtain an optimized foreground highlighting semantics guiding the route vision multi-scale feature map, wherein the formula is:

[0075]

[0076] in, represents the number of features within a class, represents the total number of eigenvalues, is the weight hyperparameter, The multi-scale feature map of the vision field that represents the foreground highlighting semantic guidance route, Representing optimized foreground highlighting semantic guidance route view multi-scale feature map; and

[0077] The optimized foreground highlighting semantics guiding route field of view multi-scale feature map is passed through the classifier to obtain the first recognition result.

[0078] That is, the cluster optimization process includes the steps of:

[0079] Determine the number of feature values ​​in a cluster set obtained by clustering the multi-scale feature map of the foreground salient semantics guiding the route field of view (for example, clustering based on the distance between feature values), that is, the number of in-class features, and subtract the number of in-class features from the total number of feature values ​​of the multi-scale feature map of the foreground salient semantics guiding the route field of view to obtain the number of out-of-class features;

[0080] Dividing the total number of feature values ​​by the number of in-class features and the number of out-class features respectively to obtain a class importance value and an out-class proportion value, and calculating the inverse of the class importance value to obtain a class constraint value;

[0081] Calculate each eigenvalue of the foreground salient semantics guiding route vision multi-scale feature map as a power function with the class importance value as an exponent, add it to an exponential value with the natural function as the base and the class constraint value as the exponent, and then multiply it by the out-of-class ratio value to obtain the foreground salient semantics guiding route vision multi-scale modulation feature map;

[0082] Performing a dot multiplication of the foreground-prominent semantics-guided route vision multi-scale feature map and the class importance value to obtain a foreground-prominent semantics-guided route vision multi-scale ontology feature map; and

[0083] The weighted sum of the foreground-prominent semantics-guided route vision multi-scale modulation feature map and the foreground-prominent semantics-guided route vision multi-scale ontology feature map is calculated with weight hyperparameters to obtain an optimized foreground-prominent semantics-guided route vision multi-scale feature map.

[0084] That is, in order to improve the obstacle target recognition accuracy of the first recognition result obtained by the classifier of the foreground highlighting semantics-guided route field of view multi-scale feature map, it is expected to improve the feature expression effect of the foreground highlighting semantics-guided route field of view multi-scale feature map based on the image deep-shallow multi-scale semantic interaction fusion and foreground highlighting representation of the first route field of view image in the complex feature representation dimension.

[0085] Based on this, in the above preferred example, while performing a clustering operation on the multi-scale feature map of the foreground salient semantics guiding the route field of view, the clustering importance measure of the feature value of the multi-scale feature map of the foreground salient semantics guiding the route field of view is combined with the regression constraint measure of the clustering operation to further perform responsive modulation with the out-of-class scale factor, and based on the clustering scaling of the feature ontology representation of the multi-scale feature map of the foreground salient semantics guiding the route field of view, the simplicity and effectiveness of the feature purification of the overall feature set of the multi-scale feature map of the foreground salient semantics guiding the route field of view by the clustering operation is established, and the inconsistency of the regression convergence logic of the probability density distribution of the image semantic generation caused by the clustering operation is suppressed, so as to enhance the model's ability to understand the generation and evolution of the manifold structure corresponding to the features under the dimension represented by the complex features during the iteration process.

[0086] In this way, the feature expression effect of the complex feature representation dimension of the multi-scale feature map of the foreground salient semantics guiding the route vision during the model iteration process is improved, thereby improving the obstacle target recognition accuracy of the first recognition result obtained by the multi-scale feature map of the foreground salient semantics guiding the route vision through the classifier. In this way, obstacle object recognition and detection in the first direction can be more accurately performed based on machine vision and artificial intelligence technology, so that the robot can make corresponding path planning adjustments by perceiving the external environment, thereby more effectively avoiding obstacles and reaching the target position, improving the flexibility and intelligence of the robot's path planning.

[0087] In step 104, the obstacle target identifier based on deep learning is used to perform obstacle target identification on the second travel route field of view image to obtain a second identification result. Specifically, the second travel route field of view image is input into the obstacle target identifier based on deep learning to obtain a second identification result. For details, reference may be made to the description of using the obstacle target identifier based on deep learning to perform obstacle target identification on the first travel route field of view image to obtain a first identification result in the embodiment of the present application.

[0088] In step 105, in response to the second recognition result that there is no obstacle object, the robot is driven to move forward along the adjusted travel angle. If the second recognition result is that there is an obstacle object, the step of driving the robot to move forward along the adjusted travel angle is not performed, and the step of adjusting the travel angle can be continued, and so on.

[0089] In summary, the above scheme is adopted to collect different travel route field images in real time through the robot's camera, and introduce machine vision and artificial intelligence technology at the back end to analyze the first travel route field image, so as to determine whether there is an obstacle object in the first direction. When an obstacle is detected, the travel angle is adjusted, and obstacle target detection in other directions is performed again. If no obstacle is detected, the robot is controlled to move forward along the travel angle. In this way, the robot can make corresponding path planning adjustments by sensing the external environment, so as to avoid obstacles and reach the target position more effectively, thereby improving the flexibility and intelligence of the robot's path planning.

[0090] Figure 2 FIG. 1 is a block diagram of a robot path planning system based on machine vision according to an exemplary embodiment. Figure 2 As shown, the robot path planning system 200 includes:

[0091] A first traveling route visual field image acquisition module 201, used to acquire a first traveling route visual field image collected by a camera of the robot;

[0092] A first obstacle target recognition module 202, configured to use a deep learning-based obstacle target recognizer to perform obstacle target recognition on the first traveling route field image to obtain a first recognition result, wherein the first recognition result is used to indicate whether an obstacle object exists;

[0093] A second traveling route field of view image acquisition module 203, used for adjusting the traveling angle of the robot and acquiring a second traveling route field of view image through the camera again;

[0094] A second obstacle target recognition module 204 is used to use the deep learning-based obstacle target identifier to perform obstacle target recognition on the second traveling route field image to obtain a second recognition result; and

[0095] The robot driving control module 205 is used to drive the robot to move forward along the adjusted travel angle in response to the second recognition result that there is no obstacle object.

[0096] Reference below Figure 3 , which shows a schematic diagram of the structure of an electronic device 600 suitable for implementing the embodiment of the present disclosure. The terminal device in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0097] like Figure 3 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0098] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0099] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0100] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0101] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0102] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0103] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0104] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0105] The modules involved in the embodiments described in the present disclosure may be implemented by software or hardware. The name of a module does not limit the module itself in some cases. For example, a test parameter acquisition module may also be described as a "module for acquiring device test parameters corresponding to a target device".

[0106] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0107] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0108] Figure 4 FIG. 1 is an application scenario diagram of a robot path planning method based on machine vision according to an exemplary embodiment. Figure 4 As shown, in this application scenario, first, a first route field image (for example, Figure 4 Then, the first route field image is input to a server (for example, Figure 4 In S) shown in , the server is capable of processing the first travel route field image based on a robot path planning algorithm based on machine vision, so as to drive the robot to move forward along the adjusted travel angle in response to the second recognition result that there is no obstacle object.

[0109] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0110] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0111] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims. Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Claims

1. A robot path planning method based on machine vision, characterized in that: include: Acquire a first travel route field of view image captured by a camera of the robot; Using a deep learning-based obstacle target identifier to perform obstacle target identification on the first traveling route field image to obtain a first identification result, wherein the first identification result is used to indicate whether an obstacle object exists; If the first recognition result indicates that there is an obstacle object, adjusting the travel angle of the robot, and collecting a second travel route field image through the camera again; If the first recognition result indicates that there is no obstacle object, the subsequent step of adjusting the travel angle is not performed; Using the deep learning-based obstacle target identifier to perform obstacle target identification on the second traveling route field image to obtain a second identification result; as well as In response to the second recognition result that there is no obstacle object, driving the robot to move forward along the adjusted travel angle; Using a deep learning-based obstacle target identifier to perform obstacle target identification on the first traveling route field image to obtain a first identification result, where the first identification result is used to indicate whether an obstacle object exists, including: Performing brightness compensation on the first traveling route field of view image to obtain a brightness compensated first traveling route field of view image; Performing feature extraction on the first traveling route visual field image after brightness compensation by using a deep neural network to obtain a traveling route visual field shallow feature map and a traveling route visual field semantic feature map; The shallow feature map of the route vision and the semantic feature map of the route vision are passed through a speech-guided attention feature selection network to obtain a semantic-guided route vision multi-scale feature map; The semantically guided route vision multi-scale feature map is passed through a region transfer module based on an attention mechanism to obtain a foreground-highlighted semantically guided route vision multi-scale feature map; Determining the first recognition result based on the multi-scale feature map of the vision of the foreground highlighting semantic guidance route; The shallow feature map of the route vision and the semantic feature map of the route vision are passed through a speech-guided attention feature selection network to obtain a semantic-guided route vision multi-scale feature map, including: Performing global mean pooling processing on each feature matrix along the channel dimension in the route vision semantic feature map to obtain a route vision semantic pooling feature vector; Passing the route vision semantic pooling feature vector through a fully connected module to perform fully connected encoding on the route vision semantic pooling feature vector to obtain a route vision semantic fully connected feature vector; Performing a weighted fusion based on the channel dimension on the fully connected feature vector of the route vision semantics and the shallow feature map of the route vision semantics to obtain a route vision semantic multi-level fusion feature map; The route vision semantic multi-level fusion feature map and the travel route vision semantic feature map are fused to obtain the semantic guidance travel route vision multi-scale feature map; The process of performing global mean pooling processing on each feature matrix along the channel dimension in the route vision semantic feature map to obtain the route vision semantic pooling feature vector comprises: calculating the global mean of each feature matrix along the channel dimension in the route vision shallow feature map to obtain the route vision semantic pooling feature vector; Among them, the fully connected feature vector of the semantics of the route vision and the shallow feature map of the route vision are weightedly fused based on the channel dimension to obtain the multi-level fusion feature map of the semantics of the route vision, including: using the fully connected feature vector of the semantics of the route vision to perform weighted processing on the shallow feature map of the route vision along the channel dimension to obtain the multi-level fusion feature map of the semantics of the route vision; The step of passing the route view semantic pooling feature vector through a fully connected module to perform fully connected encoding on the route view semantic pooling feature vector to obtain a route view semantic fully connected feature vector includes: After matrix multiplication of the route vision semantic pooling feature vector and a predetermined weight coefficient matrix, the obtained feature vector and a predetermined offset vector are added according to position to obtain a modulated route vision semantic pooling feature vector; use An activation function activates the modulated route vision semantic pooling feature vector to obtain the route vision semantic fully connected feature vector; The semantically guided route vision multi-scale feature map is passed through a region transfer module based on an attention mechanism to obtain a foreground-highlighted semantically guided route vision multi-scale feature map, including: Calculating a semantically guided route visual field spatial attention feature matrix of the semantically guided route visual field multi-scale feature map; Performing masking processing on the semantically guided route visual field spatial attention feature matrix based on a predetermined threshold to obtain a masked semantically guided route visual field spatial attention feature matrix; The foreground-highlighted semantically-guided route visual field multi-scale feature map is obtained by calculating the position-wise multiplication between each feature matrix along the channel dimension in the semantically-guided route visual field multi-scale feature map and the masked semantically-guided route visual field spatial attention feature matrix.

2. The robot path planning method based on machine vision according to claim 1, characterized in that: The deep neural network is a progressive feature pyramid network.

3. The robot path planning method based on machine vision according to claim 2, characterized in that: Calculating the semantically guided route vision spatial attention feature matrix of the semantically guided route vision multi-scale feature map includes: Performing global mean pooling processing on the semantically guided route field of view multi-scale feature map along the channel dimension to obtain a semantically guided route field of view global mean pooling feature matrix; The global mean pooling feature matrix of the semantically guided route field of view is input into the Sigmoid activation unit to obtain the spatial attention feature matrix of the semantically guided route field of view.

4. The robot path planning method based on machine vision according to claim 3, characterized in that: The method of performing masking processing on the semantic guidance route visual field spatial attention feature matrix based on a predetermined threshold to obtain a masked semantic guidance route visual field spatial attention feature matrix comprises: In response to each position feature value in the semantic guidance route visual field space attention feature matrix being greater than or equal to the predetermined threshold, the original value of the position feature value is taken, otherwise it is set to zero.

5. The robot path planning method based on machine vision according to claim 4, characterized in that: Determining the first recognition result based on the multi-scale feature map of the foreground-prominent semantics-guided route field of view includes: passing the multi-scale feature map of the foreground-prominent semantics-guided route field of view through a classifier to obtain the first recognition result.