A robot visual positioning and navigation system integrating CNN model
Features are extracted through binocular cameras and VGG models, feature points are marked, computer robot locations, and routes are planned, and the existing navigation depends on external devices is solved, achieving low-cost and high-precision autonomous navigation and feature alignment accuracy.
Patent Information
- Application Number
- CN202111276389.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-10-29
AI Technical Summary
The existing navigation methods rely on external devices, have high cost and limited accuracy, especially lidar navigation, and the application of convolutional neural networks in image recognition has the problem of insufficient feature extraction accuracy.
Photos are taken using a binocular camera, and after noise reduction and grayscale processing, the VGG model is used to extract features, and compared with the feature library, mark feature points, use distance array computer robot location, plan routes, and update the feature library using communication module to achieve Autonomous navigation.
It realizes independent navigation on a single machine, reduces peripheral costs, improves navigation accuracy and feature comparison accuracy, and can automatically plan routes within the scene to adapt to changes in occlusions.
Smart Images

Figure CN114004887B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent navigation technology, and in particular to a robot vision positioning navigation system integrating a CNN model. Background Art
[0002] Navigation is a field of study that focuses on monitoring and controlling the process of moving a process or vehicle from one place to another. Most existing navigation methods use satellite navigation, which requires the use of external equipment to achieve navigation operations.
[0003] According to the way of self-navigation, there are odometer method and lidar navigation method. However, the cumulative error of odometer method is large, and lidar navigation has high accuracy, but the cost is high. Therefore, a camera is used to capture images and the path is planned by recognition method. In image recognition, convolutional neural networks (CNN) are often used. They are a type of feedforward neural networks (CNN) that include convolution calculations and have a deep structure. They are one of the representative algorithms of deep learning. When acquiring features, the new deep convolutional neural network (VGG) developed by researchers from the Visual Geometry Group of the University of Oxford and Google DeepMind is often used. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] In view of the shortcomings of the prior art, the present invention provides a robot vision positioning and navigation system integrating a CNN model, which solves the problem that the prior art navigation relies on external devices.
[0006] (II) Technical solution
[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a robot visual positioning and navigation system integrating a CNN model, comprising the following contents:
[0008] Step 1: The binocular camera takes photos, and the photos are processed by noise reduction and grayscale;
[0009] Step 2: In the CNN model feature selection, select the VGG model, obtain features from the processed image, and compare the features with the feature library to obtain the type of feature body;
[0010] Step 3: Mark the contour on the feature body to obtain the feature points, use the characteristics of the binocular camera to measure the distance between the robot and the feature points, and obtain the distance array L;
[0011] Step 4: Calculate the positional relationship between the feature bodies through the array L, determine the position of the robot based on the position of the feature bodies, and then plan the route based on the position of the target point;
[0012] Step 5: The drive module runs according to the planned route, and uses the turning point of the planned route as the dividing point. When the robot drives to the next turning point, the drive module stops running.
[0013] Step 6: The feedback module is used to record the displacement of the driving module;
[0014] Step 7: Update the current position of the robot through the displacement information recorded by the feedback module, and then return to step 1.
[0015] Preferably, the positioning and navigation system also includes an initial position and a target position of the robot, and the robot establishes a Cartesian coordinate system with the initial position as the coordinate origin.
[0016] Preferably, the robot is also provided with a communication module. In step 2, the obtained features are preferentially uploaded to a feature library provided in the terminal for comparison. If the robot is offline, its own feature library is called.
[0017] Preferably, when the robot is connected to the Internet, the compared feature information is compared based on the connection, and the compared information is created or updated in its own feature library.
[0018] Preferably, the feature points marked in step three are selected from vertices of the feature body and equally divided points of the contour line.
[0019] Preferably, the route planning in step five should adopt the principle of proximity. When encountering an obstruction, turn left first. If there is an obstruction on the left, turn right. If there is an obstruction on the right, retreat and recalculate the planned route.
[0020] (III) Beneficial effects
[0021] The present invention provides a robot vision positioning and navigation system integrating a CNN model. It has the following beneficial effects:
[0022] 1. The present invention can be used as a single machine and automatically plans the forward route through the feature bodies in the scene.
[0023] 2. The present invention adopts a binocular camera, and the cost of using peripherals for navigation is low. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a control flow chart of the present invention. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] Example:
[0027] like Figure 1 As shown, an embodiment of the present invention provides a robot visual positioning and navigation system integrating a CNN model, including the following contents:
[0028] Step 1: Take photos with a binocular camera. A conventional binocular camera is used. The distance of a specific point can be measured by the binocular camera. The photos are processed by noise reduction and grayscale to facilitate feature extraction. When grayscale processing is performed, thirty-two levels are used to improve the accuracy of feature extraction.
[0029] Step 2: In the CNN model feature selection, select the VGG model. VGG is a convolutional neural network used for image feature extraction. Get features from the processed image, compare the features with the feature library, and obtain the type of the feature entity, that is, determine whether the feature entity is a box, wall, ground, or other entity. Through the entity type, use artificial intelligence to infer the shape features of the feature entity and its distribution in space, which is conducive to the later route planning.
[0030] Step 3: Mark the contour on the feature body to obtain feature points. Use the characteristics of the binocular camera to measure the distance between the robot and the feature points to obtain the distance array L. Use the distance array for analysis and calculation. The marked feature points select the vertices of the feature body and the equal points of the contour line. According to actual needs, determine the number of feature points selected for the same feature body. The parameters can be set manually. By selecting different numbers, a balance can be achieved between navigation accuracy and calculation amount.
[0031] The positional relationship between the feature bodies is calculated through the array L, and the position of the robot is determined by the position of the feature body, and then the route is planned according to the position of the target point;
[0032] Step 5: The driving module runs according to the planned route, and uses the turning point of the planned route as the dividing point, that is, the turning point where the planned route changes from a straight line to a curve and from a curve to a straight line. The robot drives to the next turning point, and the driving module stops running. The route planning should adopt the principle of proximity. When encountering an obstruction, the left turn is given priority. If there is an obstruction on the left, turn right. If there is an obstruction on the right, retreat and recalculate the planned route.
[0033] Step 6: The feedback module is used to record the displacement of the drive module and compensate for the operation error of the motion module itself. That is, after the motion module generates displacement, the displacement error caused by its own transmission error is compensated. After the robot position is determined in step 4, the actual displacement and the expected position distance are calculated for compensation.
[0034] Step 7: Update the current position of the robot through the displacement information recorded by the feedback module. The displacement information includes not only the position information but also the direction of displacement. Then return to step 1.
[0035] The positioning and navigation system also includes the initial position and target position of the robot, and the robot establishes a Cartesian coordinate system with the initial position as the coordinate origin, that is, determines the direction in which the robot moves.
[0036] The robot is also provided with a communication module. In step 2, the obtained features are preferentially uploaded to the feature library of the terminal for comparison. If the robot calls its own feature library when it is offline, that is, the robot can be used as a stand-alone machine, when the robot is connected to the Internet, the feature information is compared based on the network, and the compared information is created or updated in its own feature library to improve the feature comparison accuracy.
[0037] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A robot visual positioning and navigation system integrating CNN model, It is characterized in that Includes the following: Step 1: The binocular camera takes photos, and the photos are processed by noise reduction and grayscale; Step 2: In the CNN model feature selection, select the VGG model, obtain features from the processed image, and compare the features with the feature library to obtain the type of feature body; Step 3: Mark the contour on the feature body to obtain the feature points, use the characteristics of the binocular camera to measure the distance between the robot and the feature points, and obtain the distance array L; Step 4: Calculate the positional relationship between the feature bodies through the array L, determine the position of the robot based on the position of the feature bodies, and then plan the route based on the position of the target point; Step 5: The drive module runs according to the planned route, and uses the turning point of the planned route as the dividing point. When the robot drives to the next turning point, the drive module stops running. Step 6: The feedback module is used to record the displacement of the driving module; Step 7: Update the current position of the robot through the displacement information recorded by the feedback module, and then return to step 1.
2. A robot vision positioning and navigation system integrating a CNN model according to claim 1, It is characterized in that The positioning and navigation system also includes an initial position and a target position of the robot, and the robot establishes a Cartesian coordinate system with the initial position as the coordinate origin.
3. The robot vision positioning and navigation system integrating the CNN model according to claim 1, Features: The robot is also provided with a communication module. In step 2, the obtained features are preferentially uploaded to the feature library provided in the terminal for comparison. If the robot is offline, its own feature library is called.
4. A robot vision positioning and navigation system integrating a CNN model according to claim 3, Features: When the robot is connected to the Internet, the compared feature information is compared based on the connection, and the compared information is created or updated in its own feature library.
5. The robot vision positioning and navigation system integrating the CNN model according to claim 1, Features: The feature points marked in step 3 are selected from the vertices of the feature body and the equally divided points of the contour line.
6. The robot vision positioning and navigation system integrating the CNN model according to claim 1, Features: The route planning in step 5 should adopt the principle of proximity. When encountering an obstruction, turn left first. If there is an obstruction on the left, turn right. If there is an obstruction on the right, retreat and recalculate the planned route.
Citation Information
Patent Citations
Spherical robot visual positioning system
CN108332755A
Robot indoor navigation control method
CN110069058A
Indoor mobile robot visual positioning method based on convolutional neural network
CN113129373A
Method, apparatus, and system for location correction based on feature point correspondence
US20200132472A1
Mobile control method, mobile robot and computer storage medium
US20210138657A1
