Visual positioning technology and system based on scene perception and deep learning

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a visual positioning technology and system based on scene perception and deep learning, including collecting image data, data preprocessing, deep learning model training, scene feature extraction, designing visual positioning algorithms, optimizing algorithm positioning accuracy, system integration and testing, and application deployment and verification. The visual positioning technology and system based on scene perception and deep learning can significantly improve positioning accuracy and meet the needs of various application scenarios through the optimization of deep learning models and algorithms. The system can adapt to different lighting conditions, occlusion situations and dynamic environments, and has strong robustness. Through feature extraction and algorithm optimization, it can reduce the amount of calculation and improve the real-time performance and response speed of the system. With the continuous optimization of deep learning models and the development of hardware technology, the hardware cost of the system is expected to gradually decrease.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a visual positioning technology and system based on scene perception and deep learning. Background Art

[0002] Scene perception technology refers to obtaining information about the surrounding environment through sensors, cameras and other devices, and processing and analyzing this information to achieve understanding and recognition of the scene. In the visual positioning system, scene perception technology is mainly used to obtain visual information such as images and videos of the target scene, and provide data support for subsequent positioning algorithms. Deep learning is an important technology in the field of artificial intelligence. It aims to simulate the neural network in the human brain to solve complex problems. Deep learning technology extracts and transforms the input data layer by layer by building a multi-layer neural network model, and finally realizes tasks such as data classification and recognition. In the visual positioning system, deep learning technology is mainly used to extract features from visual information such as images and videos and train positioning algorithms. The background technology of visual positioning technology and system based on scene perception and deep learning involves scene perception, deep learning, computer vision, multi-sensor fusion, positioning algorithms and technologies, and related hardware and software support. The combination and application of these technologies provide strong support for the implementation of the visual positioning system. For this reason, this application now proposes a visual positioning technology and system based on scene perception and deep learning. Summary of the invention

[0003] Technical Solution To achieve the above purpose, the present invention provides the following technical solutions: a visual positioning technology and system based on scene perception and deep learning, including collecting image data, data preprocessing, deep learning model training, scene feature extraction, designing visual positioning algorithm, optimizing algorithm positioning accuracy, system integration and testing, application deployment and verification, the specific steps are as follows: S1. Use a high-resolution camera or other image acquisition device to collect a large amount of image data in the target scene; S2, performing denoising and enhancement preprocessing operations on the collected image data; S3, using the preprocessed image data and its annotation information to train the deep learning model; S4, using the trained deep learning model to extract key features from the input image; S5. Design visual positioning algorithm based on the extracted scene features; S6. Verify the positioning accuracy of the algorithm through experiments and find out the source of error; S7, integrating the optimized visual positioning algorithm into the system to form a complete visual positioning system; S8. Integrate the system into actual application scenarios, verify the positioning accuracy and reliability of the system in actual applications, and make further adjustments and optimizations based on feedback.

[0004] Preferably, the steps of collecting image data are as follows: S101. Select an image acquisition device according to the application scenario and positioning accuracy requirements; S102, adjusting the exposure and white balance parameters of the camera; S103, selecting representative scenes for image collection, so that the deep learning model can learn rich scene features; S104: Execute image acquisition operation according to preset acquisition parameters and scenes.

[0005] Preferably, the steps of data preprocessing are as follows: S201, using a mean filter and a Gaussian filter algorithm to remove noise points in the image; S202: Adjust the brightness and contrast parameters of the image, and use a histogram equalization method to improve the grayscale distribution of the image.

[0006] Preferably, the deep learning model training includes selecting a deep learning architecture, preparing training data, building a deep learning model, training a deep learning model and optimizing a deep learning model. The specific implementation includes selecting a deep learning architecture according to the application scenario and positioning requirements, collecting a large amount of image data related to the positioning task and annotating them, building a corresponding model structure according to the selected deep learning architecture, and optimizing the model according to the verification results.

[0007] Preferably, the scene feature extraction method includes traditional methods and deep learning methods. Before the rise of deep learning, scene feature extraction mainly relied on manually designed feature extraction algorithms. Compared with traditional methods, deep learning methods can automatically extract features without the need for manually designed feature extraction rules, and have stronger generalization capabilities for different scenes and objects.

[0008] Preferably, when designing the visual positioning algorithm, image data is first collected and annotated, the data containing different scenes and lighting conditions, features are extracted from the image using a deep learning model, the extracted features are matched with features in a pre-built database, and the position of the target object or scene in the image is calculated based on the matching results.

[0009] Preferably, actual scene data is collected during verification in the application deployment and verification, including images, videos, and sensor data. The positioning accuracy of the visual positioning technology is verified using labeled data sets. The accuracy of the algorithm is evaluated by comparing the difference between the position predicted by the algorithm and the actual position. The robustness of the visual positioning technology is verified in complex scenes with different lighting conditions, occlusion situations, and changes in viewing angles. The stability and reliability of the algorithm in different scenes are evaluated to ensure that it can maintain good performance in actual applications.

[0010] Compared with the prior art, the present invention provides a visual positioning technology and system based on scene perception and deep learning, which has the following beneficial effects: This visual positioning technology and system based on scene perception and deep learning can significantly improve positioning accuracy and meet the needs of various application scenarios through the optimization of deep learning models and algorithms. The system can adapt to different lighting conditions, occlusion situations and dynamic environments, and has strong robustness. Through feature extraction and algorithm optimization, it can reduce the amount of calculation and improve the real-time and response speed of the system. With the continuous optimization of deep learning models and the development of hardware technology, the hardware cost of the system is expected to gradually decrease. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the flow of the visual positioning technology method of the present invention. DETAILED DESCRIPTION

[0012] 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.

[0013] Embodiment 1: A visual positioning technology and system based on scene perception and deep learning, including image data collection, data preprocessing, deep learning model training, scene feature extraction, visual positioning algorithm design, algorithm positioning accuracy optimization, system integration and testing, application deployment and verification. The specific steps are as follows: S1. Use a high-resolution camera or other image acquisition device to collect a large amount of image data in the target scene; S2, performing denoising and enhancement preprocessing operations on the collected image data; S3, using the preprocessed image data and its annotation information to train the deep learning model; S4, using the trained deep learning model to extract key features from the input image; S5. Design visual positioning algorithm based on the extracted scene features; S6. Verify the positioning accuracy of the algorithm through experiments and find out the source of error; S7, integrating the optimized visual positioning algorithm into the system to form a complete visual positioning system; S8. Integrate the system into actual application scenarios, verify the positioning accuracy and reliability of the system in actual applications, and make further adjustments and optimizations based on feedback.

[0014] Embodiment 2: According to the first embodiment, a visual positioning technology and system based on scene perception and deep learning are proposed. The steps of collecting image data are as follows: S101. Select an image acquisition device according to the application scenario and positioning accuracy requirements; S102, adjusting the exposure and white balance parameters of the camera; S103, selecting representative scenes for image collection, so that the deep learning model can learn rich scene features; S104, performing image acquisition operations according to preset acquisition parameters and scenes; Collecting image data is an important part of visual positioning technology and systems based on scene perception and deep learning. By selecting appropriate image acquisition equipment, setting reasonable acquisition parameters, determining representative acquisition scenes and performing high-quality image acquisition operations, a solid foundation can be laid for subsequent deep learning model training and visual positioning algorithm design.

[0015] Embodiment three: According to the first embodiment, a visual positioning technology and system based on scene perception and deep learning are proposed. The steps of data preprocessing are as follows: S201, using a mean filter and a Gaussian filter algorithm to remove noise points in the image; S202, adjusting the brightness and contrast parameters of the image, and using a histogram equalization method to improve the grayscale distribution of the image; Data preprocessing is one of the key steps in visual positioning technology and systems based on scene perception and deep learning. Through reasonable preprocessing operations, the quality of image data can be improved, the feature extraction capability can be enhanced, and the model training efficiency can be improved, thus providing a solid foundation for the subsequent application of visual positioning algorithms.

[0016] Embodiment 4: According to the first embodiment, a visual positioning technology and system based on scene perception and deep learning are proposed. The deep learning model training includes selecting a deep learning architecture, preparing training data, building a deep learning model, training a deep learning model and optimizing a deep learning model. The specific implementation is to select a deep learning architecture according to the application scenario and positioning requirements, collect a large amount of image data related to the positioning task and annotate it, build a corresponding model structure according to the selected deep learning architecture, build a corresponding model structure according to the selected deep learning architecture, and optimize the model according to the verification result. Deep learning model training is one of the key links in the visual positioning technology and system based on scene perception and deep learning. By selecting a suitable deep learning architecture, preparing high-quality training data, building a reasonable model structure, and using effective optimization algorithms and regularization techniques, a deep learning model with excellent performance can be trained, providing a solid foundation for the subsequent application of visual positioning algorithms.

[0017] Embodiment five: According to the first embodiment, a visual positioning technology and system based on scene perception and deep learning are proposed. The scene feature extraction method includes a traditional method and a deep learning method. Before the rise of deep learning, scene feature extraction mainly relied on manually designed feature extraction algorithms. Compared with the traditional method, the deep learning method can automatically extract features without manually designing feature extraction rules, and has stronger generalization ability for different scenes and objects. When designing the visual positioning algorithm, image data is first collected and annotated, and the data includes different scenes and lighting conditions. A deep learning model is used to extract features from the image, and the extracted features are matched with features in a pre-built database. According to the matching results, the position of the target object or scene in the image is calculated.

[0018] Embodiment six: According to the first embodiment, a visual positioning technology and system based on scene perception and deep learning are proposed. During application deployment and verification, actual scene data, including images, videos, and sensor data, are collected. The positioning accuracy of the visual positioning technology is verified using a labeled data set. The accuracy of the algorithm is evaluated by comparing the difference between the position predicted by the algorithm and the actual position. The robustness of the visual positioning technology is verified in complex scenes with different lighting conditions, occlusion conditions, and changes in viewing angles. The stability and reliability of the algorithm in different scenes are evaluated to ensure that it can maintain good performance in actual applications.

[0019] 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 visual positioning technology and system based on scene perception and deep learning, characterized by: It includes collecting image data, data preprocessing, deep learning model training, scene feature extraction, designing visual positioning algorithms, optimizing algorithm positioning accuracy, system integration and testing, and application deployment and verification. The specific steps are as follows: S1. Use a high-resolution camera or other image acquisition device to collect a large amount of image data in the target scene; S2, performing denoising and enhancement preprocessing operations on the collected image data; S3, using the preprocessed image data and its annotation information to train the deep learning model; S4, using the trained deep learning model to extract key features from the input image; S5. Design visual positioning algorithm based on the extracted scene features; S6. Verify the positioning accuracy of the algorithm through experiments and find out the source of error; S7, integrating the optimized visual positioning algorithm into the system to form a complete visual positioning system; S8. Integrate the system into actual application scenarios, verify the positioning accuracy and reliability of the system in actual applications, and make further adjustments and optimizations based on feedback.

2. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: The steps of collecting image data are as follows: S101. Select an image acquisition device according to the application scenario and positioning accuracy requirements; S102, adjusting the exposure and white balance parameters of the camera; S103, selecting representative scenes for image collection, so that the deep learning model can learn rich scene features; S104: Execute image acquisition operation according to preset acquisition parameters and scenes.

3. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: The steps of data preprocessing are as follows: S201, using a mean filter and a Gaussian filter algorithm to remove noise points in the image; S202: Adjust the brightness and contrast parameters of the image, and use a histogram equalization method to improve the grayscale distribution of the image.

4. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: The deep learning model training includes selecting a deep learning architecture, preparing training data, building a deep learning model, training a deep learning model and optimizing a deep learning model. The specific implementation includes selecting a deep learning architecture according to the application scenario and positioning requirements, collecting image data related to the positioning task and annotating it, building a corresponding model structure according to the deep learning architecture, building a corresponding model structure according to the selected deep learning architecture, and optimizing the model according to the verification results.

5. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: The scene feature extraction method includes traditional methods and deep learning methods. Before the rise of deep learning, scene feature extraction relied on manually designed feature extraction algorithms. Compared with traditional methods, deep learning methods can automatically extract features without manually designing feature extraction rules, and have stronger generalization capabilities for different scenes and objects.

6. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: When designing the visual positioning algorithm, image data is first collected and annotated, and the data contains different scenes and lighting conditions. A deep learning model is used to extract features from the image, and the extracted features are matched with features in a pre-built database. Based on the matching results, the position of the target object or scene in the image is calculated.

7. The visual positioning technology and system based on scene perception and deep learning according to claim 1, characterized in that: During the application deployment and verification, actual scene data is collected, including images, videos, and sensor data. The positioning accuracy of the visual positioning technology is verified using labeled data sets. The accuracy of the algorithm is evaluated by comparing the difference between the position predicted by the algorithm and the actual position. The robustness of the visual positioning technology is verified in complex scenes with different lighting conditions, occlusion situations, and changes in viewing angles, and the stability and reliability of the algorithm in different scenes are evaluated.

Images