WO2020186774A1 - 基于图像检测的定位方法、装置、设备和存储介质 - Google Patents

基于图像检测的定位方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2020186774A1
WO2020186774A1 PCT/CN2019/116478 CN2019116478W WO2020186774A1 WO 2020186774 A1 WO2020186774 A1 WO 2020186774A1 CN 2019116478 W CN2019116478 W CN 2019116478W WO 2020186774 A1 WO2020186774 A1 WO 2020186774A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
vector
initial
preset
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/116478
Other languages
English (en)
French (fr)
Inventor
何维
陈佳
胡家义
杨昭
王威
钟亮
邹永卫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020186774A1 publication Critical patent/WO2020186774A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/70Determining position or orientation of objects or cameras
    • G06T7/73Determining position or orientation of objects or cameras using feature-based methods

Definitions

  • This application relates to the field of image detection, and in particular to a positioning method, device, device and storage medium based on image detection.
  • GPS Global Positioning System, Global Positioning System
  • GPS Global Positioning System
  • GPS positioning technology cannot be used for accurate positioning, so that the user cannot determine his current specific location indoors; in addition, GPS has some disadvantages of its own, such as power consumption, GPS module The cost is extremely high, the positioning accuracy is poor in rainy days, it cannot be used on devices without an integrated GPS module and cannot be passively positioned. In order to avoid the weakness of GPS positioning, achieving accurate positioning has become a technical problem that needs to be solved urgently.
  • the main purpose of this application is to provide a positioning method, device, equipment and storage medium based on image detection, aiming to realize image position recognition for accurate positioning through image detection.
  • the positioning method based on image detection includes the following steps:
  • the present application also provides a positioning device based on image detection, and the positioning device based on image detection includes:
  • a request receiving module configured to receive an image position recognition request, and obtain an initial image corresponding to the image position recognition
  • a vector conversion module configured to obtain image feature information corresponding to the initial image, and calculate an image feature vector corresponding to the initial image according to the image feature information
  • An image matching module for querying a preset image database, obtaining a target image vector matching the image feature vector, and a target image corresponding to the target image vector;
  • the position determining module is configured to determine the target position corresponding to the initial image according to the position information associated with the target image.
  • this application also provides a positioning device based on image detection
  • the image detection-based positioning device includes: a memory, a processor, and computer-readable instructions stored on the memory and running on the processor, wherein:
  • this application also provides a computer storage medium
  • the computer storage medium stores computer readable instructions, and when the computer readable instructions are executed by a processor, the steps of the positioning method based on image detection as described above are realized.
  • the embodiment of the application proposes a positioning method, device, equipment and storage medium based on image detection.
  • the user triggers the image location recognition request based on the image on the terminal.
  • the server obtains the initial image corresponding to the image location recognition request, and then the server extracts the image corresponding to the initial image by detecting the initial image.
  • the server converts the image feature information into image feature vectors, and uses the image feature vectors for image matching and image location recognition to ensure positioning accuracy.
  • the positioning in this application does not need to rely on GPS modules or other external Setting reduces the cost of positioning and improves the convenience of positioning.
  • FIG. 1 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application;
  • FIG. 2 is a schematic flowchart of a first embodiment of a positioning method based on image detection in this application;
  • FIG. 3 is a schematic diagram of an embodiment of a positioning device based on image detection in this application.
  • Figure 1 is the server of the hardware operating environment involved in the solution of the embodiments of the present application (also called the positioning device based on image detection, where the positioning device based on image detection can be a separate positioning device based on image detection
  • the structure can also be formed by combining other devices with a positioning device based on image detection).
  • the server in the embodiment of the present application refers to a computer that manages resources and provides services for users, and is generally divided into a file server, a database server, and an application-readable instruction server.
  • the computer or computer system running the above software is also called a server.
  • the server may include: a processor 1001, such as a central processing unit (Central Processing Unit, CPU), network interface 1004, user interface 1003, memory 1005, communication bus 1002, chipset, disk system, network and other hardware.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as WIreless-FIdelity, WIFI interface).
  • the memory 1005 may be a high-speed random access memory (random access memory, RAM), or stable memory (non-volatile memory), such as disk storage.
  • the memory 1005 may also be a storage device independent of the foregoing processor 1001.
  • the computer software product is stored in a storage medium (storage medium: also called computer storage medium, computer medium, readable medium, readable storage medium, computer readable storage medium, or directly called medium, etc., storage medium
  • storage medium can be a non-volatile readable storage medium, such as RAM, magnetic disk, optical disk, and includes several instructions to make a terminal device (can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute this application
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and computer-readable instructions.
  • the network interface 1004 is mainly used to connect to the back-end database and perform data communication with the back-end database;
  • the user interface 1003 is mainly used to connect to the client (the client, also called the user terminal or the terminal, the embodiment of the application
  • the terminal can be a fixed terminal or a mobile terminal, such as smart air conditioners with networking functions, smart lights, smart power supplies, smart speakers, self-driving cars, PCs, smart phones, tablets, e-book readers, portable computers, etc., terminals It contains sensors such as light sensors, motion sensors and other sensors, which will not be repeated here), to communicate with the client; and the processor 1001 can be used to call the computer-readable instructions stored in the memory 1005 and execute the following The steps in the image detection-based positioning method provided by the embodiment.
  • This embodiment proposes a positioning method based on image detection, which is applied to the server shown in FIG. 1.
  • the image detection-based positioning method includes:
  • Step S10 Receive an image location recognition request, and obtain an initial image corresponding to the image location recognition.
  • the server receives the image location recognition request. After the server receives the image location recognition request, the server obtains the initial image corresponding to the image location recognition request; the image location recognition request received by the server can be triggered in different ways, for example, a user application
  • the terminal uploads the photo version of the real street corner image, and triggers the image location recognition request based on the uploaded photo version to take the real street corner image.
  • the terminal sends the image location recognition request to the server.
  • the server After the server receives the image location recognition request, the server sends the photo version of the real street corner image As the initial image corresponding to the image location recognition request; for another example, the user selects a building map on the terminal and voice inputs "location recognition" to trigger the image location recognition request, the terminal sends the image location recognition request to the server, and the server receives After the image location recognition request, the server uses the building map selected by the user as the initial image corresponding to the image location recognition request. It should be noted that the number of initial images may be one or more. In this embodiment, one initial image is taken as an example for description.
  • Step S20 Obtain image feature information corresponding to the initial image, and calculate an image feature vector corresponding to the initial image according to the image feature information.
  • the server After the server obtains the initial image to be recognized, the server performs image detection on the initial image.
  • the image detection of the initial image by the server is implemented through an image recognition model (or understood as a deep neural network), that is, the server is preset Image recognition model.
  • the image recognition model removes the redundant information in the initial image, and then divides the initial image into regions.
  • the server obtains the image feature information corresponding to the initial image.
  • the image feature information includes the color feature, texture feature, shape feature and See the hierarchical features and so on.
  • the server converts the image feature information into an image feature vector. That is, in this embodiment, the server converts the initial image into a quantized image feature vector, so that the server can match according to the image feature vector, specifically:
  • Step S30 comparing the image feature vector with an image vector in a preset image database, and obtaining a target image vector matching the image feature vector and a target image corresponding to the target image vector.
  • the server obtains a target image similar to the initial image according to the image feature vector, specifically:
  • Step a query a preset image database, calculate the cosine value of the image feature vector and each image vector in the preset image database, and compare the cosine value with a preset cosine threshold;
  • Step b Obtain a target cosine value greater than the preset cosine threshold and a target image vector corresponding to the target cosine value, and use the preset image corresponding to the target image vector as a target image.
  • the server queries the preset image database (the preset image database is a preset database that stores each recognized image, for example, the preset image database stores the preset image "xxx street", the preset image and the image
  • the image vector is stored in association with the location information of the image).
  • the server obtains the preset image in the preset image database and the image vector corresponding to each preset image.
  • the server calculates the difference between the image feature vector and each image vector in the preset image database.
  • Cosine value that is, in this embodiment, the cosine value is used to characterize the image similarity. If the cosine value of the image feature vector and the image vector is large, the initial image corresponding to the image feature vector is similar to the preset image corresponding to the image vector, and vice versa.
  • the server compares the calculated cosine value with a preset cosine threshold, where the preset cosine threshold is a preset cosine critical value.
  • the preset cosine threshold can be set to 0.8, and the server obtains the image feature vector cosine value greater than the preset cosine threshold. Set the target image vector of the cosine threshold, and use the preset image corresponding to the target image vector as the target image.
  • Step S40 Determine the target position corresponding to the initial image according to the position information associated with the target image.
  • the server obtains the location information associated with the target image, and determines the initial image location information according to the location information of the target image. For example, if the server obtains 5 target images, the server obtains the corresponding location information of the target image, that is, the location information corresponds to the location. The middle position of the 5 positions is used as the target position corresponding to the initial image.
  • the user triggers the image location recognition request based on the image on the terminal.
  • the server obtains the initial image corresponding to the image location recognition request.
  • the server extracts the image corresponding to the initial image by detecting the initial image.
  • the server converts the image feature information into image feature vectors, and uses the image feature vectors for image matching and image location recognition to ensure positioning accuracy.
  • the positioning in this application does not need to rely on GPS modules or other The external setting reduces the positioning cost and improves the convenience of positioning.
  • This embodiment is a refinement of step S20 of the first embodiment.
  • an implementation method for converting an initial image into a corresponding image feature vector is given, and specifically includes:
  • the first image recognition model is used to preprocess the initial image. Before the first image recognition model is used to recognize the initial image, the terminal needs to construct the first image recognition model. The construction of the first image recognition model is described in this embodiment. The steps include:
  • Step S01 Collect different types of first sample images from the network, and summarize each of the first sample images to obtain a first sample image set.
  • the server collects different types of real scene images, indoor images, etc. from the network.
  • the server takes the collected real scene images and indoor images as the first sample image.
  • the server summarizes the first sample images to obtain the first sample image set.
  • the first sample image set is used to train the first image recognition model.
  • the terminal collects a large number of first sample images from the network, for example, The number of the first sample images in the sample image collection can be 50,000.
  • Step S02 Extract a first sample image of a preset ratio from the first sample image set, and construct an initial image recognition model from the first sample image of the preset ratio.
  • the server extracts the first sample with a preset ratio (wherein, the preset ratio refers to the preset extraction ratio, and the preset ratio can be flexibly set according to the scene, for example, the preset ratio is set to 0.1%) from the first sample image collection Image, the server constructs an initial image recognition model using a first sample image of a preset ratio, where the essence of the initial image recognition model is a multi-point classifier, that is, the user first annotates image feature information in the first sample image Then, the server fits a function according to the image feature information marked by the user, and the server performs a recognition classification on the image feature information in the image according to the function.
  • the preset ratio refers to the preset extraction ratio, and the preset ratio can be flexibly set according to the scene, for example, the preset ratio is set to 0.15%
  • Step S03 Train the initial image recognition model through an iterative algorithm to obtain optimal model parameters, and use the image recognition model corresponding to the optimal model parameters obtained through training as the first image recognition model.
  • the server trains the initial image recognition model through an iterative algorithm, that is, the initial image recognition model constructed by the server may have recognition errors.
  • the server uses the first image in the first sample image set.
  • the sample images train the initial image recognition model.
  • the training of the initial image recognition model is a machine learning process.
  • the machine learning process can be achieved through neural networks, genetic algorithms, support vector machines and other methods.
  • the server obtains the optimal response from the machine learning image recognition model. Model parameters, the server uses the image recognition model corresponding to the optimal model parameters obtained through training as the preset image recognition model.
  • Method one is to set the recognition parameters. For example, the recognition accuracy rate is 98%. After the server is trained, the image recognition model recognition accuracy rate reaches 98%, and the server stops. Training; Method 2: Set the number of training times in the server: 1000 times, and then the server extracts 0.1% of the first sample image from the first sample image set to train the initial image recognition model and adjust the image recognition model Parameters, iterate 1000 times, the server uses the parameters corresponding to the image recognition model after 1000 times as the optimal model parameters, and the server uses the image recognition model corresponding to the optimal model parameters obtained through training as the preset image recognition model.
  • setting a corresponding image recognition model according to specific recognition requirements can ensure the accuracy of image recognition and improve image processing efficiency.
  • the server After obtaining the first image recognition model, the server processes the initial image through the first image recognition model, specifically:
  • Step S21 Preprocess the initial image through the first image recognition model, and extract image feature information corresponding to the preprocessed initial image.
  • the server inputs the initial image into the first image recognition model.
  • the first image recognition model preprocesses the initial image. Preprocessing means that the server removes noise and noise interference by filtering, and then performs processing on the processed initial image. Thresholding the region and value thresholding the image, wherein the method of region thresholding is preferably: determining the division threshold according to the statistical characteristics of the pixel gray value of each pixel level field in the front image, and binarizing the input image to obtain two Value image; then filter out the noise pixels in the binary image through morphological erosion operation, and use morphological expansion operation to fill in weakly connected areas, and finally extract the image feature information in the binary image.
  • Step S22 Query a preset binary tree structure, construct a Huffman search tree corresponding to the image feature information, and obtain node path information of the Huffman search tree to obtain an image feature vector corresponding to the initial image.
  • the server After the server obtains the image feature information, the server queries a preset binary tree structure, where the preset binary tree structure is a tree structure set in the server according to the image information in advance, and the server queries the preset binary tree structure, and constructs the Harbin corresponding to the image feature information. Fuman searches the tree, obtains the node path information of the Huffman search tree, and obtains the image feature vector corresponding to the initial image.
  • a deep neural network is used to extract all photos or incremental photos from the environment (season, weather and other interference factors) information, extract location-related features for fitting analysis, generate a first image recognition model, and identify based on the first image
  • the model constructs a Huffman search tree to obtain the image feature vector corresponding to the initial image, so that the server can search along the tree to greatly reduce the load of the search engine.
  • This embodiment is a refinement of step S20 of the first embodiment.
  • This embodiment presents another implementation of converting the initial image into a corresponding image feature vector.
  • the difference between this embodiment and the second embodiment lies in this embodiment.
  • the second image recognition model in the example is different from the first image recognition model in the second embodiment.
  • the number of sample images used for model training is different, and the image feature vector conversion method is different. Specifically, it includes:
  • the initial image is processed by the second image recognition model.
  • the terminal Before using the second image recognition model to recognize the initial image, the terminal needs to construct a second image recognition model.
  • the steps for constructing the second image recognition model are described, including:
  • Step S04 Convert the initial image into an image mapping vector according to a preset vector mapping method, and compare the image mapping vector with each image vector in the preset image database.
  • the preset vector mapping method in the server where the preset vector mapping method refers to a preset image-to-vector mapping algorithm, for example, the preset vector mapping method can be: local sensitive hash algorithm or SIFT algorithm (SIFT algorithm, full name Scale-invariant feature transform is also called scale-invariant feature transformation algorithm), the server transforms the initial image into the corresponding image mapping vector through the preset vector mapping method.
  • SIFT algorithm full name Scale-invariant feature transform is also called scale-invariant feature transformation algorithm
  • the server After the server obtains the image mapping vector, the server compares the image mapping vector with each image vector in the preset image database (the preset image database is the same as the preset image database in the first embodiment, and will not be repeated in this embodiment) Yes, that is, the server calculates the similarity between the image mapping vector and each image vector, specifically:
  • Step S05 Obtain a similar image vector matching the image mapping vector and a similar image corresponding to the similar image vector, and use each of the similar images as a second sample image.
  • the server obtains a similar image vector matching the image mapping vector, where the matching with the image mapping vector refers to a similar image corresponding to the similar image vector with a pointing amount similarity higher than 50%.
  • the preset vector mapping method is used to convert the initial image into an image mapping vector, and the image mapping vector is used to screen some similar images as the training samples of the second image recognition mode, so as to construct an accurate recognition model through a small amount of image information ,specifically:
  • Step S06 Extract a second sample image of a preset ratio, and construct an initial image recognition model from the second sample image.
  • the server extracts a second sample image with a preset ratio from the second sample image (where the preset ratio refers to the preset extraction ratio, and the preset ratio can be flexibly set according to specific scenarios, for example, the preset ratio is set to 1%) ,
  • the server constructs the initial image recognition model by using the second sample image of the preset ratio.
  • step S07 the initial image recognition model is trained through an iterative algorithm to obtain optimal model parameters, and the image recognition model corresponding to the optimal model parameters obtained through training is used as the second image recognition model.
  • the server trains the initial image recognition model through an iterative algorithm, that is, the initial image recognition model constructed by the server may have recognition errors.
  • the server uses the second sample in the second sample image set
  • the image trains the initial image recognition model.
  • the training of the initial image recognition model is a machine learning process.
  • the machine learning process can be achieved through neural networks, genetic algorithms, support vector machines and other methods.
  • the server obtains the optimal response from the machine learning image recognition model. Model parameters, the server uses the image recognition model corresponding to the optimal model parameters obtained through training as the preset image recognition model.
  • the initial image recognition is trained through the second sample image, and the steps to obtain the second image recognition model are the same as those in the above-mentioned second embodiment, which will not be repeated in this embodiment.
  • the server uses a small amount of AND Similar images with similar initial images are used as second sample images, and the second image recognition model is trained to reduce the amount of data and ensure image processing efficiency.
  • the server uses the second image recognition model to process the initial image, specifically:
  • Step S23 Preprocess the initial image through the second image recognition model, and extract image feature information of the preprocessed initial image.
  • the server preprocesses the initial image through the second image recognition model (the preprocessing steps are the same as in the second embodiment, and will not be repeated in this embodiment), and the second image recognition model extracts the preprocessed image of the initial image Feature information, where the image feature information includes: color feature, texture feature, shape feature, and spatial relationship feature.
  • Step S24 Set the weights of the color feature, the texture feature, the shape feature, and the spatial relationship feature according to a preset search mapping table, and construct an initial vector corresponding to the image feature information.
  • the server searches the mapping table according to the preset (the preset search mapping table is a preset image search weight table, for example, the color feature weight in the preset search mapping table is 10%, the texture feature weight is 20%, the shape feature 40%, and the space The relationship feature weight is 30%.
  • the weight value in the preset search mapping table can be set according to the retrieval situation. For example, the retrieved picture is a street corner, and the image contains trees. The color of the trees will change with the seasons. In order to increase the retrieval accuracy, The server reduces the weight value of the color) and sets the weight of the color feature, texture feature, shape feature, and spatial relationship feature to obtain the initial vector corresponding to the image feature information.
  • Step S25 Perform normalization processing on the initial vector to obtain an image feature vector corresponding to the initial image.
  • the server performs normalization processing on the initial vector, where the vector normalization processing usually refers to reducing the dimension of the vector to obtain the image feature vector corresponding to the initial image.
  • the initial image is subjected to vector mapping to obtain the image mapping vector, and the similar image of the initial image is obtained from the preset image database according to the image mapping vector.
  • the server uses the similar image as the second sample image to construct the first image using the deep learning model.
  • the second image recognition model is to transform the initial image into the corresponding image feature vector through the second image recognition model.
  • the image detection-based positioning method includes:
  • Step S41 Obtain the target image and the position information associated with the target image, determine the position corresponding to the position information, and determine whether the distance between the positions exceeds a preset distance threshold.
  • the server obtains the target image and the position information associated with the target image, and the server determines the corresponding position of the position information.
  • the number of target images obtained in this application is variable, and it can be one or more.
  • the server outputs the target image
  • the corresponding position is used as the target position corresponding to the initial image.
  • the server needs to determine the target position corresponding to an initial image according to each position, specifically:
  • the server obtains each location and calculates the distance between each location.
  • the server compares the distance between each location with a preset distance threshold to finally determine the target location corresponding to the initial image.
  • the preset distance threshold can be more specific Flexible setting of the scene, for example, when the initial image is an indoor image, the preset distance threshold can be set to 10m, and when the initial image is an outdoor image, the preset distance threshold can be set to 100m.
  • Step S42 When the distance between the positions does not exceed a preset distance threshold, calculate the center position of each position, and use the center position as the target position corresponding to the initial image.
  • the server calculates the center position of each position and uses the center position as the target position corresponding to the initial image, for example, target image a corresponds to position a, and target image b corresponds to position b, the target image c corresponds to position c; the distance between position a and position b is 50 meters, the distance between position a and position c is 30 meters; the distance between position b and position c is 40 meters, and the preset distance threshold is 100m, the server determines that the distance between each position does not exceed the preset distance threshold, the server calculates the center distance of position a, position b, and position c, that is, the server makes a circumscribed circle of the corresponding points of position a, position b, and position c, The server uses the center of the circumscribed circle as the center positions of position a, position b, and position c, that is, in this embodiment, the terminal uses the midpoint of position a and position b
  • the server sends the position corresponding to each target image to the corresponding terminal, so that the user can select the corresponding initial image.
  • the target location That is, the server cannot accurately locate the corresponding position of the initial image, and the server outputs each position for the user to choose, which increases the flexibility of positioning.
  • the server may perform a real-time display according to the determined target location, and specifically includes:
  • Step S50 Receive a real-scene display request, obtain relevant information of the display position corresponding to the real-scene display request, and convert the relevant information into real-scene image output.
  • the user can trigger a real-scene display request on the terminal, and the terminal sends the real-scene display request to the server.
  • the server receives the real-scene display request
  • the server obtains the picture data associated with the display position, and the server uses the picture data to generate a three-dimensional real-scene environment Material, through SLAM (simultaneous Localization and mapping) and other real-time map drawing methods can provide users with a real-world map display experience and improve user experience.
  • SLAM simultaneous Localization and mapping
  • this embodiment can be combined with the fourth embodiment. Specifically, when the terminal receives multiple locations sent by the server, and the server cannot determine the corresponding location of the initial image, the user triggers a real-scene display request on the terminal , The server displays the real image corresponding to each target image for the user to confirm the target location corresponding to the initial image.
  • an embodiment of the present application also proposes a positioning device based on image detection, and the positioning device based on image detection includes:
  • the request receiving module 10 is configured to receive an image position recognition request, and obtain an initial image corresponding to the image position recognition;
  • the vector conversion module 20 is configured to obtain image feature information corresponding to the initial image, and calculate an image feature vector corresponding to the initial image according to the image feature information;
  • the image matching module 30 is configured to query a preset image database, obtain a target image vector matching the image feature vector, and a target image corresponding to the target image vector;
  • the position determining module 40 is configured to determine the target position corresponding to the initial image according to the position information associated with the target image.
  • the positioning device based on image detection includes:
  • the sample collection module is configured to collect different types of first sample images from the network, and collect each of the first sample images to obtain a first sample image set;
  • the first construction module is configured to extract a first sample image of a preset ratio from the first sample image set, and construct an initial image recognition model from the first sample image of the preset ratio;
  • the first generation module is configured to train the initial image recognition model through an iterative algorithm to obtain optimal model parameters, and use the image recognition model corresponding to the optimal model parameters obtained through training as the first image recognition model.
  • the vector conversion module 20 includes:
  • a first processing unit configured to preprocess the initial image through the first image recognition model, and extract image feature information corresponding to the preprocessed initial image
  • the first query unit is configured to query a preset binary tree structure, construct a Huffman search tree corresponding to the image feature information, and obtain node path information of the Huffman search tree to obtain the image feature corresponding to the initial image vector.
  • the positioning device based on image detection includes:
  • a mapping conversion module configured to convert the initial image into an image mapping vector according to a preset vector mapping method, and compare the image mapping vector with each image vector in the preset image database;
  • a sample acquisition module configured to acquire a similar image vector matching the image mapping vector, and a similar image corresponding to the similar image vector, and use each of the similar images as a second sample image;
  • the second construction module is used to extract a second sample image of a preset ratio, and construct an initial image recognition model from the second sample image;
  • the second generation module is used to train the initial image recognition model through an iterative algorithm to obtain optimal model parameters, and use the image recognition model corresponding to the optimal model parameters obtained through training as the second image recognition model.
  • the vector conversion module 20 includes:
  • the second processing unit is configured to preprocess the initial image through the second image recognition model, and extract image feature information of the preprocessed initial image, wherein the image feature information includes: color features, Texture characteristics, shape characteristics and spatial relationship characteristics;
  • the second query unit is configured to set the weights of the color feature, the texture feature, the shape feature, and the spatial relationship feature according to a preset search mapping table, and construct an initial vector corresponding to the image feature information;
  • the normalization processing unit is configured to perform normalization processing on the initial vector to obtain an image feature vector corresponding to the initial image.
  • the image matching module 30 includes:
  • the cosine calculation unit is configured to query a preset image database, calculate the cosine value of the image feature vector and each image vector in the preset image database, and compare the cosine value with a preset cosine threshold;
  • the image determining unit is configured to obtain a target cosine value greater than the preset cosine threshold and a target image vector corresponding to the target cosine value, and use the preset image corresponding to the target image vector as a target image.
  • the position determining module 40 includes:
  • the position range determining unit is configured to obtain the target image and the position information associated with the target image, determine the position corresponding to the position information, and determine whether the distance between the positions exceeds a preset distance threshold;
  • the position determining unit is configured to calculate the center position of each position when the distance between the positions does not exceed a preset distance threshold, and use the center position as the target position corresponding to the initial image.
  • the steps implemented by the functional modules of the image detection-based positioning device can refer to the various embodiments of the image detection-based positioning method of the present application, which will not be repeated here.
  • the embodiment of the present application also proposes a computer storage medium.
  • the computer storage medium stores computer readable instructions, and when the computer readable instructions are executed by the processor, the operations in the image detection-based positioning method provided in the foregoing embodiments are implemented.
  • the computer storage medium may be a non-volatile computer readable storage medium.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Library & Information Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

一种基于图像检测的定位方法、定位装置、设备和存储介质,涉及图像检测领域,该定位方法包括:接收图像位置识别请求,获取图像位置识别对应的初始图像(S10);获取初始图像对应的图像特征信息,并根据图像特征信息计算初始图像对应的图像特征向量(S20);将图像特征向量与预设图像数据库中的图像向量进行比对,获取与图像特征向量匹配的目标图像向量,及目标图像向量对应的目标图像(S30);根据目标图像关联的位置信息,确定初始图像对应的目标位置(S40)。该定位方法通过对初始图像检测提取图像特征信息,并根据图像特征信息对应的图像特征向量进行图像匹配,实现了准确地图像定位,降低了定位成本。

Description

基于图像检测的定位方法、装置、设备和存储介质
本申请要求于2019年03月16日提交中国专利局、申请号为201910205775.5、发明名称为“基于图像检测的定位方法、装置、设备和存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及图像检测领域,尤其涉及基于图像检测的定位方法、装置、设备和存储介质。
背景技术
当前的定位主要是通过GPS(Global Positioning System,全球定位系统)实现的,但是GPS在常常会受到地理位置等因素的限制。
例如,在用户处于室内环境时,采用GPS定位技术并不能进行准确地定位,这样用户无法确定自己当前在室内的具体位置;此外,GPS还有一些自身的缺点,如,耗电、GPS模组成本极高,阴雨天定位精度差,无法在没有集成GPS模块的设备上使用和不能被动定位等问题。为了避免GPS定位的弱点,实现准确地定位成为了当前亟待解决的技术问题。
发明内容
本申请的主要目的在于提供一种基于图像检测的定位方法、装置、设备和存储介质,旨在通过图像检测,实现图像位置识别进行准确定位。
为实现上述目的,本申请提供基于图像检测的定位方法,所述基于图像检测的定位方法包括以下步骤:
接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
此外,为实现上述目的,本申请还提供一种基于图像检测的定位装置,所述基于图像检测的定位装置包括:
请求接收模块,用于接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
向量转化模块,用于获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
图像匹配模块,用于查询预设图像数据库,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
位置确定模块,用于根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
此外,为实现上述目的,本申请还提供一种基于图像检测的定位设备;
所述基于图像检测的定位设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,其中:
所述计算机可读指令被所述处理器执行时实现如上所述的基于图像检测的定位方法的步骤。
此外,为实现上述目的,本申请还提供计算机存储介质;
所述计算机存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如上述的基于图像检测的定位方法的步骤。
本申请实施例提出的一种基于图像检测的定位方法、装置、设备和存储介质。本申请实施例中用户在终端上基于图像触发图像位置识别请求,服务器接收到图像位置识别请求之后,服务器获取图像位置识别请求对应的初始图像,然后,服务器通过对初始图像检测提取初始图像对应的图像特征信息,服务器将图像特征信息转化为图像特征向量,并利用图像特征向量进行图像匹配,进行图像位置识别,保证了定位准确性,同时由于本申请中的定位不需要依赖GPS模块或者其他外部设置,降低了定位成本,提高了定位的便捷性。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的装置结构示意图;
图2为本申请基于图像检测的定位方法第一实施例的流程示意图;
图3为本申请基于图像检测的定位装置一实施例的示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
如图1所示,图1是本申请实施例方案涉及的硬件运行环境的服务器(又叫基于图像检测的定位设备,其中,基于图像检测的定位设备可以是由单独的基于图像检测的定位装置构成,也可以是由其他装置与基于图像检测的定位装置组合形成)结构示意图。
本申请实施例服务器指一个管理资源并为用户提供服务的计算机,通常分为文件服务器、数据库服务器和应用可读指令服务器。运行以上软件的计算机或计算机系统也被称为服务器。相对于普通PC(personal computer)个人计算机来说,服务器在稳定性、安全性、性能等方面都要求较高;如图1所示,该服务器可以包括:处理器1001,例如中央处理器(Central Processing Unit,CPU),网络接口1004,用户接口1003,存储器1005,通信总线1002、芯片组、磁盘系统、网络等硬件等。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如无线保真WIreless-FIdelity,WIFI接口)。存储器1005可以是高速随机存取存储器(random access memory,RAM),也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
如图1所示,该计算机软件产品存储在一个存储介质(存储介质:又叫计算机存储介质、计算机介质、可读介质、可读存储介质、计算机可读存储介质或者直接叫介质等,存储介质可以是非易失性可读存储介质,如RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法,作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及计算机可读指令。
在图1所示的服务器中,网络接口1004主要用于连接后台数据库,与后台数据库进行数据通信;用户接口1003主要用于连接客户端(客户端,又叫用户端或终端,本申请实施例终端可以固定终端也可以是移动终端,如,带联网功能的智能空调、智能电灯、智能电源、智能音箱、自动驾驶汽车、PC、智能手机、平板电脑、电子书阅读器、便携计算机等,终端中包含传感器比如光传感器、运动传感器以及其他传感器,在此不再赘述),与客户端进行数据通信;而处理器1001可以用于调用存储器1005中存储的计算机可读指令,并执行本申请以下实施例提供的基于图像检测的定位方法中的步骤。
本实施例提出了一种基于图像检测的定位方法,应用于如图1所示的服务器。
参照图2,在本申请基于图像检测的定位方法的第一实施例中,所述基于图像检测的定位方法包括:
步骤S10,接收图像位置识别请求,获取所述图像位置识别对应的初始图像。
服务器接收图像位置识别请求,在服务器接收到图像位置识别请求之后,服务器获取图像位置识别请求对应的初始图像;其中,服务器接收到的图像位置识别请求可以是不同的方式触发的,例如,用户应用终端上传拍照版街角实景图像,并基于上传的拍照版拍街角实景图像触发图像位置识别请求,终端将图像位置识别请求发送至服务器,服务器接收到图像位置识别请求之后,服务器将拍照版街角实景图像作为图像位置识别请求对应的初始图像;再比如,用户在终端上选择一张建筑物图,并语音输入“位置识别”触发图像位置识别请求,终端将图像位置识别请求发送至服务器,服务器接收到图像位置识别请求之后,服务器将用户选择的建筑物图作为图像位置识别请求对应的初始图像。需要说明的是,初始图像数量可以是一张或者多张,本实施例中以一张初始图像为例进行说明。
步骤S20,获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量。
在服务器获取到待识别的初始图像之后,服务器对初始图像进行图像检测,其中,服务器对初始图像的图像检测是通过图像识别模型(或者理解为深度神经网络)实现的,即,服务器中预先设置图像识别模型,图像识别模型去除初始图像中的冗余信息,然后,对初始图像进行区域划分,服务器获取初始图像对应的图像特征信息,图像特征信息包括图像的颜色特征、纹理特征、形状特征和空见层次特征等等。在服务器获取到初始图像对应的图像特征信息之后,服务器对图像特征信息转化为图像特征向量。即,在本实施例中服务器将初始图像转化为量化的图像特征向量,以使服务器根据图像特征向量进行匹配,具体地:
步骤S30,将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像。
服务器根据图像特征向量,获取初始图像相似的目标图像,具体地:
步骤a,查询预设图像数据库,计算所述图像特征向量与预设图像数据库中的各图像向量的余弦值,将所述余弦值与预设余弦阈值进行比较;
步骤b,获取大于所述预设余弦阈值的目标余弦值,及所述目标余弦值对应的目标图像向量,将所述目标图像向量对应的预设图像作为目标图像。
即,服务器查询预设图像数据库(预设图像数据库为预先设置的保存有各个已识别图像的数据库,例如,预设图像数据库中保存有预设图像“xxx街道”,预设图像与该图像的图像向量和该图像的位置信息关联存储),服务器获取预设图像数据库中的预设图像,和各个预设图像对应的图像向量,服务器计算图像特征向量与预设图像数据库中的各图像向量的余弦值;即,本实施例中用余弦值表征图像相似度,若图像特征向量与图像向量的余弦值大,则图像特征向量对应初始图像与图像向量对应预设图像相近,反之。
服务器将计算得到的余弦值与预设余弦阈值进行比较,其中,预设余弦阈值为预先设置的余弦临界值,例如,预设余弦阈值可以设置为0.8,服务器获取与图像特征向量余弦值大于预设余弦阈值的目标图像向量,并将目标图像向量对应的预设图像作为目标图像。
步骤S40,根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
服务器获取目标图像关联的位置信息,并根据目标图像的位置信息确定初始图像位置信息,例如,服务器获取到5个目标图像,服务器获取目标图像的对应位置信息,即位置信息对应为位置,服务器将5个位置的中间位置作为初始图像对应的目标位置。
在本实施例中用户在终端上基于图像触发图像位置识别请求,服务器接收到图像位置识别请求之后,服务器获取图像位置识别请求对应的初始图像,然后,服务器通过对初始图像检测提取初始图像对应的图像特征信息,服务器将图像特征信息转化为图像特征向量,并利用图像特征向量进行图像匹配,进行图像位置识别,保证了定位准确性,同的由于本申请中的定位不需要依赖GPS模块或者其他外部设置,降低了定位成本,提高了定位的便捷性。
进一步地,在本申请第一实施例的基础上,提出了本申请基于图像检测的定位方法的第二实施例。
本实施例是第一实施例步骤S20的细化,本实施例中给出了一种将初始图像转化为对应的图像特征向量的实现方式,具体地,包括:
本实施例通过第一图像识别模型对初始图像预处理,在利用第一图像识别模型识别初始图像之前,终端需要先构建第一图像识别模型,本实施例中说明了第一图像识别模型的构建步骤,包括:
步骤S01,从网络上采集不同类型的第一样本图像,并将各所述第一样本图像汇总得到第一样本图像集合。
服务器从网络上采集不同类型的实景图像、室内图像等,服务器将采集的实景图像、室内图像作为第一样本图像,服务器将各第一样本图像汇总得到第一样本图像集合,需要说明的是,第一样本图像集合是用来训练第一图像识别模型的,为了保证第一图像识别模型识别的准确性,终端从网络上采集的第一样本图像数量较多,例如,第一样本图像集合中第一样本图像的数量可以为50000张。
步骤S02,从所述第一样本图像集合中抽取预设比例的第一样本图像,并通过预设比例的所述第一样本图像构建初始图像识别模型。
服务器从第一样本图像集合中抽取预设比例(其中,预设比例是指预先设置抽取比例,预设比例可以根据场景灵活设置,例如,预设比例设置为0.1%)的第一样本图像,服务器利用预设比例的第一样本图像构建初始图像识别模型,其中,该初始图像识别模型的本质是一个多分的分类器,即,用户首先在第一样本图像中标注图像特征信息,然后,服务器根据用户标注的图像特征信息拟合一个函数,服务器根据该函数对图像中图像特征信息进行一个识别分类。
步骤S03,通过迭代算法训练所述初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第一图像识别模型。
服务器通过迭代算法训练初始图像识别模型,即,服务器构建的初始图像识别模型可能存在识别错误等的情况,为了提高图像识别模型识别的准确性,服务器利用该第一样本图像集合中的第一样本图像对初始图像识别模型进行训练。其中,对初始图像识别模型训练是一个机器学习的过程,该机器学习的过程可以是通过神经网络、遗传算法、支持向量机等多种方式实现,服务器通过机器学习图像识别模型对应的得到最优模型参数,服务器并将训练得到最优模型参数对应的图像识别模型作为预设图像识别模型。
本实施例中对初始图像识别模型的训练可以采用两种方式,方式一,设置识别参数,例如,识别准确率为98%,在服务器训练到图像识别模型识别的准确率到达98%,服务器停止训练;方式二,服务器中设置训练次数:为1000次,然后,服务器每次从第一样本图像集合中抽取0.1%的第一样本图像对初始图像识别模型进行训练,调整图像识别模型的参数,迭代1000次,服务器将1000次之后的图像识别模型对应的参数作为最优模型参数,并服务器将训练得到最优模型参数对应的图像识别模型作为预设图像识别模型。在本实施例中根据具体识别需求设置对应的图像识别模型,可以保证图像识别的准确性,提高图像处理效率。
在得到第一图像识别模型之后,服务器通过第一图像识别模型对初始图像进行处理,具体地:
步骤S21,通过所述第一图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像对应的图像特征信息。
服务器将初始图像输入至第一图像识别模型中,第一图像识别模型对初始图像进行预处理,预处理是指服务器通过滤波处理的方法去除的噪声和噪点干扰,然后对处理后的初始图像进行区域阈值化,值得阈值化图像,其中区域阈值化方法优选为:根据所述前方图像中每个像素水平领域的像素灰度值的统计特征确定划分阈值,对输入图像进行二值化处理得到二值图像;再通过形态学腐蚀运算滤除二值图像中的噪点像素,并运用形态学膨胀运算填补弱连接区域,最后提取所述二值图像中的图像特征信息。
步骤S22,查询预设二叉树结构,构建所述图像特征信息对应的哈夫曼搜索树,并获取所述哈夫曼搜索树的节点路径信息,得到所述初始图像对应的图像特征向量。
在服务器获取到图像特征信息之后,服务器查询预设二叉树结构,其中,预设二叉树结构为服务器中预先根据图像信息设置的树形结构,服务器查询预设二叉树结构,并构建图像特征信息对应的哈夫曼搜索树,并获取哈夫曼搜索树的节点路径信息,得到初始图像对应的图像特征向量。
本实施例中通过深度神经网络将所有照片或增量照片抽离环境(季节、天气等干扰因素)信息,抽取位置相关特征做拟合分析,生成第一图像识别模型,并基于第一图像识别模型构建哈夫曼搜索树,得到初始图像对应的图像特征向量,使得服务器沿树型搜索可以极大减少搜索引擎负载。
进一步地,在本申请第一实施例的基础上,提出了本申请基于图像检测的定位方法的第三实施例。
本实施例是第一实施例步骤S20的细化,本实施例给出了另一种将初始图像转化为对应的图像特征向量的实现方式,本实施例与第二实施例的区别在于本实施例中第二图像识别模型与第二实施例中的第一图像识别模型,用于模型训练的样本图像数量不同,图像特征向量转化方式不同,具体地,包括:
本实施例中通过所述第二图像识别模型对初始图像处理。在利用第二图像识别模型识别初始图像之前,终端需要先构建第二图像识别模型,本实施例中说明了第二图像识别模型的构建步骤,包括:
步骤S04,将所述初始图像按预设向量映射法转化为图像映射向量,并将所述图像映射向量与预设图像数据库中的各图像向量进行比对。
服务器中预设向量映射法,其中,预设向量映射法是指预先设置的图像到向量映射算法,例如,预设向量映射法可以是:局部敏感哈希算法或 SIFT算法(SIFT算法,全称Scale-invariant feature transform又叫尺度不变特征转换算法),服务器通过预设的向量映射法将初始图像转化为对应的图像映射向量。
在服务器获取到图像映射向量之后,服务器将图像映射向量与预设图像数据库(预设图像数据库与第一实施例的中预设图像数据库相同,本实施例不作赘述)中的各图像向量进行比对,即,服务器计算图像映射向量与各个图像向量的相似度,具体地:
步骤S05,获取与所述图像映射向量匹配的相似图像向量,及所述相似图像向量对应的相似图像,将各所述相似图像作为第二样本图像。
服务器获取与图像映射向量匹配的相似图像向量,其中,与图像映射向量匹配是指向量相似度高于50%,及相似图像向量对应的相似图像。本实施例中利用预设向量映射法,将初始图像转化为图像映射向量,以利用图像映射向量筛选部分相似图像作为第二图像识别模式的训练样本,以通过少量的图像信息构建准确的识别模型,具体地:
步骤S06,抽取预设比例的第二样本图像,并通过所述第二样本图像构建初始图像识别模型。
服务器从第二样本图像中抽取预设比例(其中,预设比例是指预先设置的抽取比例,预设比例可以根据具体场景灵活设置,例如,预设比例设置为1%)的第二样本图像,服务器利用预设比例的第二样本图像构建初始图像识别模型。
步骤S07,通过迭代算法训练初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第二图像识别模型。
服务器通过迭代算法训练初始图像识别模型,即,服务器构建的初始图像识别模型可能存在识别错误等的情况,为了提高图像识别模型识别的准确性,服务器利用该第二样本图像集合中的第二样本图像对初始图像识别模型进行训练。其中,对初始图像识别模型训练是一个机器学习的过程,该机器学习的过程可以是通过神经网络、遗传算法、支持向量机等多种方式实现,服务器通过机器学习图像识别模型对应的得到最优模型参数,服务器并将训练得到最优模型参数对应的图像识别模型作为预设图像识别模型。
在本实施例中通过第二样本图像对初始图像识别进行训练,得到第二图像识别模型的步骤与上述第二实施例的相同,本实施例中不作赘述,本实施例中服务器利用少量的与初始图像相似的相似图像作为第二样本图像,训练得到第二图像识别模型,减少了数据数量,保证了图像的处理效率。
在得到第二图像识别模型之后,服务器利用第二图像识别模型对初始图像进行处理,具体地:
步骤S23,通过所述第二图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像的图像特征信息。
服务器通过第二图像识别模型对所述初始图像进行预处理(预处理的步骤与第二实施例相同,本实施例不作赘述),第二图像识别模型提取经过预处理的所述初始图像的图像特征信息,其中,所述图像特征信息包括:颜色特征、纹理特征,形状特征和空间关系特征。
步骤S24,按预设搜索映射表设置所述颜色特征、所述纹理特征、所述形状特征和所述空间关系特征的权重,并构建所述图像特征信息对应的初始向量。
服务器按预设搜索映射表(预设搜索映射表为预先设置的图像搜索权重表,例如,预设搜索映射表中颜色特征权重为10%、纹理特征权重为20%、形状特征40%、空间关系特征权重为30%,其中预设搜索映射表中的权重值可以根据检索情况设置,例如,检索图片为街角,图像中包含树木,树木随四季变化颜色会发生变化,为了增加检索准确性,服务器将颜色的权重值降低)设置颜色特征、纹理特征、形状特征和空间关系特征的权重,得到图像特征信息对应的初始向量。
步骤S25,将所述初始向量进行归一化处理,得到所述初始图像对应的图像特征向量。
服务器将初始向量进行归一化处理,其中,向量归一化处理通常是指降低向量维度,得到初始图像对应的图像特征向量。
本实施例中先将初始图像进行向量映射得到图像映射向量,根据图像映射向量从预设图像数据库中获取初始图像的相似图像,然后,服务器将相似图像作为第二样本图像用深度学习模型构建第二图像识别模型,以通过第二图像识别模型将初始图像转化对应的图像特征向量,本实施例中的方法在降低数据处理量的同时保证了图像识别的准确性。
进一步地,在上述实施例的基础上提出了,本申请基于图像检测的定位方法的第四实施例。
本实施例是第一实施例中步骤S40的细化,本实施例中具体说明的终端如何确定目标位置,具体地,所述基于图像检测的定位方法包括:
步骤S41,获取所述目标图像,及所述目标图像关联的位置信息,确定所述位置信息对应的位置,并判断各述位置之间的距离是否超过预设距离阈值。
服务器获取目标图像和目标图像关联的位置信息,服务器确定位置信息对应位置,其中,本申请中获取目标图像数量不定,可以一个或者是多个,在目标图像的数量是一个时,服务器输出目标图像对应的位置作为初始图像对应的目标位置,在目标图像的数量为多个时,服务器需要根据各个位置确定一个初始图像对应的目标位置,具体地:
服务器获取各个位置,并计算各个位置之间的距离,服务器将各个位置之间的距离与预先设置的距离阈值进行比较,以最终确定初始图像对应的目标位置,其中,预设距离阈值可以更加具体场景灵活设置,例如,初始图像为室内图像时,预设距离阈值可以设置为10m,初始图像为室外图像时,预设距离阈值可以设置为100m。
步骤S42,在各所述位置之间的距离不超过预设距离阈值时,计算各位置的中心位置,并将所述中心位置作为初始图像对应的目标位置。
服务器确定各个位置之间的距离不超过预设距离阈值时,服务器计算各位置的中心位置,并将中心位置作为初始图像对应的目标位置,例如,目标图像a对应位置a,目标图像b对应位置 b,目标图像c对应位置c;位置a位置b之间的距离为50米,位置a位置c之间的距离为30米;位置b位置c之间的距离为40米,预设距离阈值为100m,则服务器判定各个位置之间的距离不超过预设距离阈值,服务器计算位置a、位置b和位置c的中心距离,即,服务器作位置a、位置b和位置c对应点的外接圆,服务器将外接圆的圆心作为位置a、位置b和位置c的中心位置,即,本实施例中终端将位置a和位置b的中点作为初始图像对应的目标位置。本实施例中根据各个目标图像对应位置确定目标位置,使得目标位置计算根据准确。
需要补充说明的是,为了减小定位误差,在各位置之间的距离超过预设距离阈值时,服务器将各所述目标图像对应的位置发送至对应终端,以供用户选择所述初始图像对应的目标位置。即,服务器不可以准确地定位到初始图像对应位置,服务器将各个位置进行输出,以供用户选择,增加了定位的灵活性。
进一步地,在上述实施例的基础上提出了,本申请基于图像检测的定位方法的第五实施例。
本实施例是第一实施例中步骤S40的之后的步骤,本实施例中服务器可以根据确定的目标位置进行实景展示,具体地,包括:
步骤S50,接收实景显示请求,获取实景显示请求对应显示位置的相关信息,并将相关信息转化为实景图像输出。
即,用户可以在终端上触发实景显示请求,终端将实景显示请求发送至服务器,在服务器接收到实景显示请求之后,服务器获取显示位置关联存储的图片数据,并服务器利用该图片数据生成三维实景环境的素材,通过SLAM(simultaneous localization and mapping)等即时地图绘制手段,可以给用户提供实景地图显示的体验,提高了用户体验。
需要补充说明的是,本实施例可以与第四实施例进行结合,具体地,在终端接收到服务器发送的多个位置时,服务器不可以确定初始图像对应位置时,用户在终端触发实景显示请求,服务器显示各个目标图像对应的实景图像,以供用户确认初始图像对应的目标位置。
此外,参照图3,本申请实施例还提出了一种基于图像检测的定位装置,所述基于图像检测的定位装置包括:
请求接收模块10,用于接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
向量转化模块20,用于获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
图像匹配模块30,用于查询预设图像数据库,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
位置确定模块40,用于根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
可选地,所述的基于图像检测的定位装置,包括:
样本采集模块,用于从网络上采集不同类型的第一样本图像,并将各所述第一样本图像汇总得到第一样本图像集合;
第一构建模块,用于从所述第一样本图像集合中抽取预设比例的第一样本图像,并通过预设比例的所述第一样本图像构建初始图像识别模型;
第一生成模块,用于通过迭代算法训练所述初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第一图像识别模型。
可选地,所述向量转化模块20,包括:
第一处理单元,用于通过所述第一图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像对应的图像特征信息;
第一查询单元,用于查询预设二叉树结构,构建所述图像特征信息对应的哈夫曼搜索树,并获取所述哈夫曼搜索树的节点路径信息,得到所述初始图像对应的图像特征向量。
可选地,所述的基于图像检测的定位装置,包括:
映射转化模块,用于将所述初始图像按预设向量映射法转化为图像映射向量,并将所述图像映射向量与预设图像数据库中的各图像向量进行比对;
样本获取模块,用于获取与所述图像映射向量匹配的相似图像向量,及所述相似图像向量对应的相似图像,将各所述相似图像作为第二样本图像;
第二构建模块,用于抽取预设比例的第二样本图像,并通过所述第二样本图像构建初始图像识别模型;
第二生成模块,用于通过迭代算法训练初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第二图像识别模型。
可选地,所述向量转化模块20,包括:
第二处理单元,用于通过所述第二图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像的图像特征信息,其中,所述图像特征信息包括:颜色特征、纹理特征,形状特征和空间关系特征;
第二查询单元,用于按预设搜索映射表设置所述颜色特征、所述纹理特征、所述形状特征和所述空间关系特征的权重,并构建所述图像特征信息对应的初始向量;
归一化处理单元,用于将所述初始向量进行归一化处理,得到所述初始图像对应的图像特征向量。
可选地,所述图像匹配模块30,包括:
余弦计算单元,用于查询预设图像数据库,计算所述图像特征向量与预设图像数据库中的各图像向量的余弦值,将所述余弦值与预设余弦阈值进行比较;
图像确定单元,用于获取大于所述预设余弦阈值的目标余弦值,及所述目标余弦值对应的目标图像向量,将所述目标图像向量对应的预设图像作为目标图像。
可选地,所述位置确定模块40,包括:
位置范围确定单元,用于获取所述目标图像,及所述目标图像关联的位置信息,确定所述位置信息对应的位置,并判断各述位置之间的距离是否超过预设距离阈值;
位置确定单元,用于在各所述位置之间的距离不超过预设距离阈值时,计算各位置的中心位置,并将所述中心位置作为初始图像对应的目标位置。
其中,基于图像检测的定位装置的各个功能模块实现的步骤可参照本申请基于图像检测的定位方法的各个实施例,此处不再赘述。
此外,本申请实施例还提出一种计算机存储介质。
所述计算机存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现上述实施例提供的基于图像检测的定位方法中的操作。所述计算机存储介质可以是非易失性计算机可读存储介质。
需要说明的是,在本文中,诸如第一和第二等之类的关系术语仅仅用来将一个实体/操作/对象与另一个实体/操作/对象区分开来,而不一定要求或者暗示这些实体/操作/对象之间存在任何这种实际的关系或者顺序;术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种基于图像检测的定位方法,其特征在于,所述基于图像检测的定位方法包括以下步骤:
    接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
    获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
    将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
    根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
  2. 如权利要求1所述的基于图像检测的定位方法,其特征在于,所述接收图像位置识别请求,获取所述图像位置识别对应的初始图像的步骤之后,包括:
    从网络上采集不同类型的第一样本图像,并将各所述第一样本图像汇总得到第一样本图像集合;
    从所述第一样本图像集合中抽取预设比例的第一样本图像,并通过预设比例的所述第一样本图像构建初始图像识别模型;
    通过迭代算法训练所述初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第一图像识别模型。
  3. 如权利要求2所述的基于图像检测的定位方法,其特征在于,所述获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量的步骤,包括:
    通过所述第一图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像对应的图像特征信息;
    查询预设二叉树结构,构建所述图像特征信息对应的哈夫曼搜索树,并获取所述哈夫曼搜索树的节点路径信息,得到所述初始图像对应的图像特征向量。
  4. 如权利要求1所述的基于图像检测的定位方法,其特征在于,所述接收图像位置识别请求,获取所述图像位置识别对应的初始图像的步骤之后,包括:
    将所述初始图像按预设向量映射法转化为图像映射向量,并将所述图像映射向量与预设图像数据库中的各图像向量进行比对;
    获取与所述图像映射向量匹配的相似图像向量,及所述相似图像向量对应的相似图像,将各所述相似图像作为第二样本图像;
    抽取预设比例的第二样本图像,并通过所述第二样本图像构建初始图像识别模型;
    通过迭代算法训练初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第二图像识别模型。
  5. 如权利要求4所述的基于图像检测的定位方法,其特征在于,所述获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量的步骤,包括:
    通过所述第二图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像的图像特征信息,其中,所述图像特征信息包括:颜色特征、纹理特征,形状特征和空间关系特征;
    按预设搜索映射表设置所述颜色特征、所述纹理特征、所述形状特征和所述空间关系特征的权重,并构建所述图像特征信息对应的初始向量;
    将所述初始向量进行归一化处理,得到所述初始图像对应的图像特征向量。
  6. 如权利要求1所述的基于图像检测的定位方法,其特征在于,所述将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像的步骤,包括:
    查询预设图像数据库,计算所述图像特征向量与预设图像数据库中的各图像向量的余弦值,将所述余弦值与预设余弦阈值进行比较;
    获取大于所述预设余弦阈值的目标余弦值,及所述目标余弦值对应的目标图像向量,将所述目标图像向量对应的预设图像作为目标图像。
  7. 如权利要求1所述的基于图像检测的定位方法,其特征在于,所述根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置的步骤,包括:
    获取所述目标图像,及所述目标图像关联的位置信息,确定所述位置信息对应的位置,并判断各所述位置之间的距离是否超过预设距离阈值;
    在各所述位置之间的距离不超过预设距离阈值时,计算各位置的中心位置,并将所述中心位置作为初始图像对应的目标位置。
  8. 一种基于图像检测的定位装置,其特征在于,所述基于图像检测的定位装置包括:
    请求接收模块,用于接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
    向量转化模块,用于获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
    图像匹配模块,用于查询预设图像数据库,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
    位置确定模块,用于根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
  9. 如权利要求8所述的基于图像检测的定位装置,其特征在于,所述基于图像检测的定位装置包括:
    样本采集模块,用于从网络上采集不同类型的第一样本图像,并将各所述第一样本图像汇总得到第一样本图像集合;
    第一构建模块,用于从所述第一样本图像集合中抽取预设比例的第一样本图像,并通过预设比例的所述第一样本图像构建初始图像识别模型;
    第一生成模块,用于通过迭代算法训练所述初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第一图像识别模型。
  10. 如权利要求9所述的基于图像检测的定位装置,其特征在于,所述向量转化模块,包括:
    第一处理单元,用于通过所述第一图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像对应的图像特征信息;
    第一查询单元,用于查询预设二叉树结构,构建所述图像特征信息对应的哈夫曼搜索树,并获取所述哈夫曼搜索树的节点路径信息,得到所述初始图像对应的图像特征向量。
  11. 如权利要求8所述的基于图像检测的定位装置,其特征在于,所述的基于图像检测的定位装置,包括:
    映射转化模块,用于将所述初始图像按预设向量映射法转化为图像映射向量,并将所述图像映射向量与预设图像数据库中的各图像向量进行比对;
    样本获取模块,用于获取与所述图像映射向量匹配的相似图像向量,及所述相似图像向量对应的相似图像,将各所述相似图像作为第二样本图像;
    第二构建模块,用于抽取预设比例的第二样本图像,并通过所述第二样本图像构建初始图像识别模型;
    第二生成模块,用于通过迭代算法训练初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第二图像识别模型。
  12. 如权利要求11所述的基于图像检测的定位装置,其特征在于,所述向量转化模块,包括:
    第二处理单元,用于通过所述第二图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像的图像特征信息,其中,所述图像特征信息包括:颜色特征、纹理特征,形状特征和空间关系特征;
    第二查询单元,用于按预设搜索映射表设置所述颜色特征、所述纹理特征、所述形状特征和所述空间关系特征的权重,并构建所述图像特征信息对应的初始向量;
    归一化处理单元,用于将所述初始向量进行归一化处理,得到所述初始图像对应的图像特征向量。
  13. 如权利要求8所述的基于图像检测的定位装置,其特征在于,所述图像匹配模块,包括:
    余弦计算单元,用于查询预设图像数据库,计算所述图像特征向量与预设图像数据库中的各图像向量的余弦值,将所述余弦值与预设余弦阈值进行比较;
    图像确定单元,用于获取大于所述预设余弦阈值的目标余弦值,及所述目标余弦值对应的目标图像向量,将所述目标图像向量对应的预设图像作为目标图像。
  14. 如权利要求8所述的基于图像检测的定位装置,其特征在于,所述位置确定模块,包括:
    位置范围确定单元,用于获取所述目标图像,及所述目标图像关联的位置信息,确定所述位置信息对应的位置,并判断各述位置之间的距离是否超过预设距离阈值;
    位置确定单元,用于在各所述位置之间的距离不超过预设距离阈值时,计算各位置的中心位置,并将所述中心位置作为初始图像对应的目标位置。
  15. 一种基于图像检测的定位设备,其特征在于,所述基于图像检测的定位设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,其中:
    所述计算机可读指令被所述处理器执行时实现以下步骤:
    接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
    获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
    将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
    根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
  16. 如权利要求15所述的基于图像检测的定位设备,其特征在于,所述接收图像位置识别请求,获取所述图像位置识别对应的初始图像的步骤之后,包括:
    从网络上采集不同类型的第一样本图像,并将各所述第一样本图像汇总得到第一样本图像集合;
    从所述第一样本图像集合中抽取预设比例的第一样本图像,并通过预设比例的所述第一样本图像构建初始图像识别模型;
    通过迭代算法训练所述初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第一图像识别模型。
  17. 如权利要求16所述的基于图像检测的定位设备,其特征在于,所述获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量的步骤,包括:
    通过所述第一图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像对应的图像特征信息;
    查询预设二叉树结构,构建所述图像特征信息对应的哈夫曼搜索树,并获取所述哈夫曼搜索树的节点路径信息,得到所述初始图像对应的图像特征向量。
  18. 如权利要求15所述的基于图像检测的定位设备,其特征在于,所述接收图像位置识别请求,获取所述图像位置识别对应的初始图像的步骤之后,包括:
    将所述初始图像按预设向量映射法转化为图像映射向量,并将所述图像映射向量与预设图像数据库中的各图像向量进行比对;
    获取与所述图像映射向量匹配的相似图像向量,及所述相似图像向量对应的相似图像,将各所述相似图像作为第二样本图像;
    抽取预设比例的第二样本图像,并通过所述第二样本图像构建初始图像识别模型;
    通过迭代算法训练初始图像识别模型,得到最优模型参数,并将训练得到最优模型参数对应的图像识别模型作为第二图像识别模型。
  19. 如权利要求18所述的基于图像检测的定位设备,其特征在于,所述获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量的步骤,包括:
    通过所述第二图像识别模型对所述初始图像进行预处理,提取经过预处理的所述初始图像的图像特征信息,其中,所述图像特征信息包括:颜色特征、纹理特征,形状特征和空间关系特征;
    按预设搜索映射表设置所述颜色特征、所述纹理特征、所述形状特征和所述空间关系特征的权重,并构建所述图像特征信息对应的初始向量;
    将所述初始向量进行归一化处理,得到所述初始图像对应的图像特征向量。
  20. 一种计算机存储介质,其特征在于,所述计算机存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如以下步骤:
    接收图像位置识别请求,获取所述图像位置识别对应的初始图像;
    获取所述初始图像对应的图像特征信息,并根据所述图像特征信息计算所述初始图像对应的图像特征向量;
    将所述图像特征向量与预设图像数据库中的图像向量进行比对,获取与所述图像特征向量匹配的目标图像向量,及所述目标图像向量对应的目标图像;
    根据所述目标图像关联的位置信息,确定所述初始图像对应的目标位置。
PCT/CN2019/116478 2019-03-16 2019-11-08 基于图像检测的定位方法、装置、设备和存储介质 Ceased WO2020186774A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910205775.5 2019-03-16
CN201910205775.5A CN110070579A (zh) 2019-03-16 2019-03-16 基于图像检测的定位方法、装置、设备和存储介质

Publications (1)

Publication Number Publication Date
WO2020186774A1 true WO2020186774A1 (zh) 2020-09-24

Family

ID=67365339

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/116478 Ceased WO2020186774A1 (zh) 2019-03-16 2019-11-08 基于图像检测的定位方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN110070579A (zh)
WO (1) WO2020186774A1 (zh)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070579A (zh) * 2019-03-16 2019-07-30 平安科技(深圳)有限公司 基于图像检测的定位方法、装置、设备和存储介质
CN110751171B (zh) * 2019-09-06 2025-08-05 深圳平安医疗健康科技服务有限公司 图像数据分类方法、装置、计算机设备和存储介质
CN111340015B (zh) * 2020-02-25 2023-10-20 北京百度网讯科技有限公司 定位方法和装置
CN111324746B (zh) * 2020-02-25 2023-04-11 北京百度网讯科技有限公司 视觉定位方法、装置、电子设备和计算机可读存储介质
CN111522988B (zh) * 2020-05-29 2022-07-15 深圳市商汤科技有限公司 图像定位模型获取方法及相关装置
CN112036504B (zh) * 2020-09-15 2024-12-20 平安国际智慧城市科技股份有限公司 温度测量模型的训练方法、装置、设备及存储介质
CN114332511B (zh) * 2022-03-14 2022-05-27 新极技术(北京)有限公司 一种废钢厚度占比的识别方法
CN114820789B (zh) * 2022-04-26 2025-02-07 陕西新高斯激光科技有限公司 一种不规则目标智能识别及定位方法与系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573735A (zh) * 2015-01-05 2015-04-29 广东小天才科技有限公司 基于图像拍摄以优化定位的方法、智能终端及服务器
CN105246039A (zh) * 2015-10-20 2016-01-13 深圳大学 一种基于图像处理的室内定位方法及系统
CN106407315A (zh) * 2016-08-30 2017-02-15 长安大学 一种基于街景图像数据库的车辆自主定位方法
CN107967457A (zh) * 2017-11-27 2018-04-27 全球能源互联网研究院有限公司 一种适应视觉特征变化的地点识别与相对定位方法及系统
US20180174321A1 (en) * 2016-12-15 2018-06-21 Hyundai Motor Company Vehicle Localization Apparatus and Method
CN110070579A (zh) * 2019-03-16 2019-07-30 平安科技(深圳)有限公司 基于图像检测的定位方法、装置、设备和存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108228757A (zh) * 2017-12-21 2018-06-29 北京市商汤科技开发有限公司 图像搜索方法和装置、电子设备、存储介质、程序
CN108549899B (zh) * 2018-03-07 2022-02-15 中国银联股份有限公司 一种图像识别方法及装置
CN108960289B (zh) * 2018-06-08 2021-03-16 清华大学 医用图像分类装置及方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573735A (zh) * 2015-01-05 2015-04-29 广东小天才科技有限公司 基于图像拍摄以优化定位的方法、智能终端及服务器
CN105246039A (zh) * 2015-10-20 2016-01-13 深圳大学 一种基于图像处理的室内定位方法及系统
CN106407315A (zh) * 2016-08-30 2017-02-15 长安大学 一种基于街景图像数据库的车辆自主定位方法
US20180174321A1 (en) * 2016-12-15 2018-06-21 Hyundai Motor Company Vehicle Localization Apparatus and Method
CN107967457A (zh) * 2017-11-27 2018-04-27 全球能源互联网研究院有限公司 一种适应视觉特征变化的地点识别与相对定位方法及系统
CN110070579A (zh) * 2019-03-16 2019-07-30 平安科技(深圳)有限公司 基于图像检测的定位方法、装置、设备和存储介质

Also Published As

Publication number Publication date
CN110070579A (zh) 2019-07-30

Similar Documents

Publication Publication Date Title
WO2020186774A1 (zh) 基于图像检测的定位方法、装置、设备和存储介质
JP6139364B2 (ja) 人物特定装置、人物特定方法及びプログラム
US20210233319A1 (en) Context-aware tagging for augmented reality environments
WO2020087704A1 (zh) 信贷信息管理方法、装置、设备和存储介质
WO2021051558A1 (zh) 基于知识图谱的问答方法、装置和存储介质
US20080112621A1 (en) User interface for face recognition
WO2020019403A1 (zh) 用电量异常检测方法、装置、设备及可读存储介质
KR20190120335A (ko) 타겟 트래킹 방법, 장치, 전자기기와 기억매체
CN110929770A (zh) 一种基于图像处理的智能追踪方法、系统、设备及可读介质
WO2020186777A1 (zh) 图像检索方法、装置、设备及计算机可读存储介质
WO2011013910A2 (ko) 증강 현실을 제공하는 방법과 그를 위한 서버 및 휴대용 단말기
WO2012050268A1 (ko) 위치 기반 모바일 응용 마켓 플레이스 시스템
US11308732B2 (en) Face recognition method and apparatus
WO2020207038A1 (zh) 基于人脸识别的人数统计方法、装置、设备及存储介质
WO2011093669A2 (ko) 객체 인식시스템 및 이를 이용하는 객체 인식 방법
WO2020085558A1 (ko) 고속분석 영상처리장치 및 그 장치의 구동방법
CN111311022B (zh) 发电量预测方法、装置、设备及计算机可读存储介质
WO2020080734A1 (ko) 얼굴 인식 방법 및 얼굴 인식 장치
WO2022131727A1 (ko) 부동산 정보 제공 장치 및 부동산 정보 제공 방법
WO2014133251A1 (ko) 엘에스에이치 알고리즘의 자료조회결과의 특징점을 이용한 매칭포인트 추출시스템 및 그 방법
CN111782980A (zh) 地图兴趣点的挖掘方法、装置、设备及存储介质
WO2015037815A1 (ko) 스마트 기기 내 시맨틱 검색 시스템 및 검색방법
CN108287873A (zh) 数据处理方法及相关产品
WO2020111844A2 (ko) 물체 레이블을 활용한 비주얼 슬램에서의 영상 특징점 강화 방법 및 장치
WO2015129987A1 (ko) 사물 인식에 기반한 광고를 제공하는 서비스 장치, 사물 인식에 기반한 광고를 수신하는 사용자 장치, 사물 인식에 기반한 광고를 제공하는 시스템, 방법 및 컴퓨터 프로그램이 기록된 기록매체

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19919694

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19919694

Country of ref document: EP

Kind code of ref document: A1