A positioning method, system and storage medium for a wearable device

Abstract

This invention discloses a positioning method, system, and storage medium for a wearable device. The method includes: extracting and distinguishing global and local features of objects and histograms in a scene image, dividing the scene into different regions, and constructing feature image indexes for those regions; when positioning is required, activating the user's wearable device to capture a scene image, extracting the image features and local feature sets, and transmitting the feature sets to a central control system in real time via relay for identification; using the image feature indexes of the candidate regions, obtaining candidate locations and images of the candidate locations, and then matching them with the candidate images using the local feature sets. This invention is simple to install and configure, can achieve high-precision positioning, and has better practicality and applicability; this invention actually transmits and uses de-identified feature data, occupies low communication bandwidth, and has better privacy protection.

Description

Technical Field

[0001] This invention relates to the field of real-time positioning technology, and more specifically to a positioning method, system and storage medium for wearable devices. Background Technology

[0002] Knowing the real-time location of people in the environment is a necessary requirement for highly intelligent places such as smart homes, smart factories, and nursing homes, both now and in the future.

[0003] For example, in nursing homes, knowing the real-time location and status of each resident can significantly reduce the workload of back-end monitoring staff, allowing them to address issues immediately. In smart factories and construction sites, security levels can be set based on each worker's real-time location, providing real-time alerts and management. In a future era where service robots are ubiquitous, such as in a home environment where the homeowner is in a bedroom and the robot is in the living room, the robot proactively finding the homeowner is a common need. If the robot cannot directly determine the person's location, the problem will be extremely complex, and blindly searching will significantly impact the user experience.

[0004] Early work in this field focused on deploying numerous surveillance cameras in the environment and running human recognition algorithms to monitor and track the location of each person in real time. While this approach has been applied in places like smart unmanned supermarkets, it suffers from significant drawbacks, including high installation and construction costs, high computational complexity, privacy concerns due to comprehensive surveillance, and potential errors in individual identification. On the other hand, some nursing homes have adopted UWB (Ultra-Wideband) facilities to locate people in their rooms, but these also face challenges such as high installation costs, complex configurations, low positioning accuracy, and limited applicability on a large scale. Due to the relatively low positioning accuracy, this approach is not ideal for applications where robots serve humans.

[0005] To address the aforementioned issues, a positioning method, system, and storage medium for wearable devices are needed. Summary of the Invention

[0006] This invention discloses a positioning method for a wearable device, which includes the following steps:

[0007] Based on the extraction and differentiation of global and local features of objects and histograms in scene images, after dividing different regions, the feature image feature index of the region is constructed using the word band model of local features;

[0008] When there is a positioning requirement, the user-worn device is activated to capture scene images, extract the image features and local feature sets, and transmit the feature sets to the central control system in real time via relay for identification. The identification process is based on object and histogram features to distinguish the candidate regions where the features are located.

[0009] Using the image feature index of the candidate region, the candidate location and the image of the candidate location are obtained, and then the matching with the candidate image is completed through the local feature set.

[0010] Further, the extraction of the image features includes:

[0011] Extract the [H,S] features of the image, and find the m closest region categories among the K categories of the [H,S] features;

[0012] The ORB features of the image are extracted to obtain the positions of n feature points and their corresponding feature values.

[0013] Multiple feature values ​​are searched in BoW to find their corresponding feature words and form a feature word set;

[0014] The image is extracted and stored based on the feature word set as a candidate location image;

[0015] The position transformation matrix between the candidate position images is calculated, and the matching error is calculated. The candidate position image with the smallest matching error is used as the initial positioning point, and the corresponding shooting position is accurately calculated to obtain the real-time positioning of the user's human body.

[0016] Further, after extracting the [H,S] features and ORB features of the image, the process includes:

[0017] When both front and rear camera images have the aforementioned [H,S] features and ORB features, discard the image with fewer feature points and retain the image with the most feature points;

[0018] If the front and rear camera images do not have the aforementioned [H,S] and ORB features, they are discarded.

[0019] Furthermore, the step of searching multiple feature values ​​in BoW to find their corresponding feature words and forming a feature word set includes:

[0020] Obtain the feature word set w = {w1, w2, ..., wn};

[0021] Set a specified threshold for the number of elements in the feature word set;

[0022] If n is less than the specified threshold, the currently processed image is invalid, and the process jumps to the next frame for processing.

[0023] Further, the feature word set is extracted and stored in the image as a candidate location image, including:

[0024] Obtain the feature word set ki;

[0025] Extract the corresponding stored images that contain the element w in the feature word set ki, and calculate the parameter bi of the stored images;

[0026] The keyframes selected are sorted from largest to smallest by the bi parameter, and no more than 10 corresponding stored images are extracted as candidate location images.

[0027] Further, after extracting the corresponding stored image containing the element w in the feature word set ki, the process includes:

[0028] Calculate parameter ai, where ai is the logarithm of the stored image corresponding to the w element contained in the feature word set ki.

[0029] Preferably, the parameter bi is the ratio of the parameter ai to the total number of w elements in the feature word set ki.

[0030] Preferably, the bi parameter value corresponding to the extracted stored image is greater than 0.35.

[0031] On the other hand, another technical solution adopted by the present invention is to provide a positioning system for a wearable device.

[0032] The positioning system includes:

[0033] The image acquisition module is used to activate the front and rear dual cameras of the user-worn device to capture scene images, and transmit the images to the central control system for identification in real time via a wireless relay through a wireless module.

[0034] The feature extraction module extracts image features, obtaining the positions and corresponding feature values ​​of n feature points. It then searches the BoW algorithm for these feature values ​​to find their corresponding feature words, forming a feature word set. The feature extraction methods employed by the module include, but are not limited to, the FAST algorithm and the BRIEF algorithm.

[0035] A candidate location image generation module extracts and stores the image based on the feature word set as a candidate location image.

[0036] The real-time human body positioning module calculates the position transformation matrix between the candidate position images and calculates the matching error. It uses the candidate position image with the smallest matching error as the initial positioning point and accurately calculates the corresponding shooting position to obtain the user's real-time human body positioning.

[0037] The central control system receives and identifies the feature set in real time.

[0038] The call unit automatically acquires the real-time human location calculated by the real-time human location module, and when it receives a call command, sends a call alarm signal containing the real-time human location of the user to the property management center of the user's area.

[0039] Furthermore, the present invention also includes a client comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the above method.

[0040] Furthermore, the present invention also includes a computer-readable storage medium storing a computer program and applied to a client, characterized in that the computer program, when executed by a processor, implements the steps of the above method.

[0041] As can be seen from the above technical solution, it has at least the following advantages and positive effects:

[0042] I. Compared with existing methods, the present invention does not rely on fixed cameras or pre-deployed and configured UWB systems. It is simple to install and configure, and can achieve high-precision positioning. It has better practicality and applicability, and can better cooperate with robots to provide services.

[0043] Second, compared with existing camera monitoring methods, although the method of the present invention uses a camera, it actually transmits and uses desensitized feature data, and only turns on the video image when there is a special need. It has low communication bandwidth consumption and better privacy protection function.

[0044] Third, this invention uses bag-of-words technology for positioning, which can achieve fast positioning and query; at the same time, compared with the traditional bag-of-words technology, it combines histograms and object features in the previous layer, which can achieve regional matching and adapt to a larger scene.

[0045] Fourth, this invention directly utilizes the map constructed by the robot SLAM method as the basis to realize the positioning technology, thereby enabling a more direct and effective integration with the robot's subsequent services. Compared with traditional methods, it saves tedious work such as positioning, registration, and coordinate transformation, and has higher positioning accuracy. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0047] Figure 1 This is a schematic flowchart of a positioning method for a wearable device provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the wearable device of the present invention;

[0049] Figure 3 This is a mapping structure diagram of the location and image features of the present invention;

[0050] Figure 4 This is a flowchart illustrating a method for generating a feature word set based on feature words according to another embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of the structure of a positioning system for a wearable device according to another embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram of the call unit of the system of the present invention.

[0053] Figure 7 This is a schematic diagram of the call unit of the system of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1

[0056] Figure 1 This is a flowchart illustrating a positioning method for a wearable device according to an embodiment of the present invention. The method includes the following steps:

[0057] S1: Based on the extraction and differentiation of global and local features of objects and histograms in the scene image, after dividing different regions, the feature image feature index of the region is constructed using the word band model of local features;

[0058] S2: When there is a positioning requirement, the user-worn device is activated to capture scene images, extract the image features and local feature sets, and transmit the feature sets to the central control system in real time via relay for identification. The identification process is based on object and histogram features to distinguish the candidate regions where the features are located.

[0059] S3: Using the image feature index of the candidate region, obtain the candidate position and the image of the candidate position, and then complete the matching with the candidate image through the local feature set.

[0060] The wearable device may consist of a camera, a low-power computing unit, a power supply, a switch button, and a wireless transmission device. A preferred structure of the wearable device in this embodiment is shown below. Figure 2 As shown, the wearable device can be integrated with a customized vest. Users can wear the vest for quick and easy operation. The device also features front and rear cameras to avoid unidirectional obstruction; the battery and processor are located within easy reach for convenient button operation and replacement. The wires and data cables between components are sealed inside the vest to prevent interference.

[0061] The user dons the wearable device vest of this invention, activates the image acquisition mode of the vest, and simultaneously rides a smart navigation wheelchair with real-time positioning enabled. The user moves through the environment, collecting scene data. The scene data includes multiple sets of [P, I1, I2], where P represents the location and direction on the electronic map where the data is being collected, and I1 and I2 represent the current images captured by the front and rear cameras, respectively. All data is uploaded to the central control system for subsequent centralized processing.

[0062] For each image I in a set of multiple groups [P, I1, I2], the following features are extracted as a feature sequence:

[0063] A: Color histogram sequence based on chromaticity; H: Same as above;

[0064] B: Sequence phrases based on object detection: S (e.g., S = [door, chair, fire hydrant], indicating that a door, chair, and fire hydrant can be detected in the current image);

[0065] C: Low-level feature set based on ORB; F: Same as above;

[0066] Ultimately, a combination of {P, [H, S, F]} can be formed, constituting a mapping from location to image features. The specific method is as follows:

[0067] First, based on the combination of {P, [H, S, F]}, {[P, H, S]} is extracted to form a new combination of data. Then, K-means clustering is used to divide the user's location into K region categories. Specifically, in the clustering algorithm, for two different data points a = [Pa, Ha, Sa] and b = [Pb, Hb, Sb], the core distance calculation uses the following formula:

[0068] dis(a,b)=|Pa-Pb|+ɑ||Ha-Hb||+β(Sa,Sb)

[0069] in, This represents the distance between the coordinates of two points; ||Ha-Hb||=∑(|ha i -hbi |) represents the sum of the absolute values ​​of the differences between corresponding terms, calculated using the normalized result; (Sa,Sb)=i, where i represents the sum of the number of objects of different types in Sa and Sb. ɑ and β are coefficients, generally taken as 1.

[0070] This yields K distinct regions, categorized by image features and their locations, with the center feature of each category serving as the category center. Then, for each category's image and location, based on feature S, the Badov Words feature dictionary processing method is applied to construct K categories' ORB feature dictionaries (BoW). These dictionaries support the rapid identification of the closest image within the same category based on ORB features. Furthermore, Figure 3 This is a mapping structure diagram between location and image features.

[0071] ORB features consist of keypoints and descriptors. The feature points used in the ORB features are "OrientedFAST," and the descriptors are "RotatedBRIEF." ORB features offer a significant speed improvement over SIFT and SURF while maintaining rotation and scale invariance of the feature points. Furthermore, the ORB features can be detected and extracted using the FAST method, specifically including:

[0072] Select a point P in the image, and draw a circle with P as the center and a radius of 3 pixels. If there are n consecutive pixels on the circumference whose gray values ​​are greater or less than the gray value of point P, then P is considered a feature point. Typically, n is set to 12. To speed up feature point extraction and quickly eliminate non-feature points, first check the gray values ​​at positions 1, 9, 5, and 13. If P is a feature point, then at these four positions, three or more pixels will have gray values ​​greater than or less than the gray value of point P; otherwise, the point is directly eliminated.

[0073] The extracted ORB feature points lack directionality, and their performance deteriorates with image rotation. The gray-scale centroid method can be used to calculate the principal direction for each feature point. Specifically, the gray-scale centroid method involves: treating the neighborhood of the feature point as a window region, calculating the centroid of this window region, and finally connecting the centroid to the feature point and finding the angle between this line and the horizontal coordinate axis; this angle represents the direction of the feature point.

[0074] Example 2

[0075] Figure 4 This is a flowchart illustrating a method for generating a feature word set based on feature values ​​according to another embodiment of the present invention. The method includes the following steps:

[0076] S10: Obtain the feature word set w = {w1, w2, ..., wn};

[0077] S11: Set a specified threshold for the number of elements in the feature word set;

[0078] S12: If n is less than the specified threshold, the currently processed image is invalid, and the process jumps to the next frame image for processing.

[0079] The Bag-of-Words (BoW) model, in information retrieval, assumes that for a text, its word order and syntax are ignored, and it is simply regarded as a set of words, or a combination of words. The occurrence of each word in the text is independent and does not depend on whether other words appear. That is, any word appearing at any position in the document will not be affected by the semantics of the document.

[0080] When processing images, the image is first visually decomposed, and the visual parts that constitute basic semantic meaning are filtered and separated to obtain several visual feature parts, i.e., feature values. These feature values ​​are then searched in a bag-of-words system to obtain and summarize the visual words representing the visual feature parts. Preferably, the K-Means algorithm can be used to construct the feature word set. The K-Means algorithm is an indirect clustering method based on the similarity measure between samples. This algorithm uses K as a parameter to divide N object feature words into K clusters, so that the similarity within each cluster is high, while the similarity between clusters is low.

[0081] Example 3

[0082] Figure 5 This is a schematic diagram of a wearable device positioning system according to another embodiment of the present invention. The system includes an image acquisition module 100, a feature extraction module 200, a candidate location image generation module 300, a real-time human body positioning module 400, and a call unit 500.

[0083] The image acquisition module 100 is used to activate the front and rear dual cameras of the user's wearable device to capture scene images, and transmit the images to the client for identification in real time via wireless relay through the wireless module.

[0084] The feature extraction module 200 extracts the image features, obtains the positions of n feature points and their corresponding feature values, and searches for multiple feature values ​​in BoW to find their corresponding feature words and form a feature word set.

[0085] The candidate location image generation module 300 extracts and stores the image based on the feature word set as a candidate location image.

[0086] The real-time human body positioning module 400 calculates the position transformation matrix between the candidate position images and calculates the matching error. It uses the candidate position image with the smallest matching error as the initial positioning point and accurately calculates the corresponding shooting position to obtain the user's real-time human body positioning.

[0087] The call unit 500 automatically obtains the real-time human location calculated by the real-time human location module, and when it receives a call command, sends a call alarm signal containing the real-time human location of the user to the property management center where the user is located.

[0088] Furthermore, the method for the feature extraction module 200 to extract the image features includes:

[0089] Extract the [H,S] features of the image, and find the m closest region categories among the K categories of the [H,S] features;

[0090] The ORB features of the image are extracted to obtain the positions of n feature points and their corresponding feature values.

[0091] Further, after extracting the [H,S] features and ORB features of the image, the process includes:

[0092] When both front and rear camera images have the aforementioned [H,S] features and ORB features, discard the image with fewer feature points and retain the image with the most feature points;

[0093] If the front and rear camera images do not have the aforementioned [H,S] and ORB features, they are discarded.

[0094] Furthermore, the step of searching multiple feature values ​​in BoW to find their corresponding feature words and forming a feature word set includes:

[0095] Obtain the feature word set w = {w1, w2, ..., wn};

[0096] Set a specified threshold for the number of elements in the feature word set;

[0097] If n is less than the specified threshold, the currently processed image is invalid, and the process jumps to the next frame for processing.

[0098] Furthermore, the method by which the candidate location image generation module 300 generates candidate location images includes:

[0099] Obtain the feature word set ki;

[0100] Extract the corresponding stored images that contain the element w in the feature word set ki, and calculate the parameter bi of the stored images;

[0101] The keyframes selected are sorted from largest to smallest by the bi parameter, and no more than 10 corresponding stored images are extracted as candidate location images.

[0102] Further, after extracting the corresponding stored image containing the element w in the feature word set ki, the process includes:

[0103] Calculate parameter ai, where ai is the logarithm of the stored image corresponding to the w element contained in the feature word set ki.

[0104] Preferably, the parameter bi is the ratio of the parameter ai to the total number of w elements in the feature word set ki.

[0105] Preferably, the bi parameter value corresponding to the extracted stored image is greater than 0.35.

[0106] Furthermore, during the use of this system, the user wears a vest containing wearable devices and walks around the environment. The positioning mode can be activated via a button to inform the central control system and the robot's own location.

[0107] Users wear vests and move around in their environment, activating location mode actively or passively. User commands can include: actively sending a location request for the robot to provide service, or being detected as having fallen and needing assistance, etc. There are many types of user commands and different processing methods. This example uses actively sending a location request for the robot to provide service to illustrate the subsequent process in detail.

[0108] The wearable device's vest features dual front and rear cameras, which are activated to capture scene images. For qualified images, histogram features H, object features S, and the set of ORB feature operators F are extracted respectively.

[0109] The criteria for judging a qualified image are as follows: when extracting the set of ORB feature operators F of the image, the positions of n feature points and their corresponding feature values ​​can be obtained; images with fewer feature points are discarded (there may be occlusion); when there are images from both the front and rear cameras, at least one image is usually retained; when the features of the two images before and after are not obvious, they can be discarded and the next frame can be obtained.

[0110] The set of three features (H, S, F) and the user's instructions (such as an instruction to actively locate and request robot services) are packaged together and sent to the central control system via wireless relay. The reason for sending only features and not images is to protect customer privacy; anonymized data represented by features is used. When the central control center receives the packaged feature combination data, it performs parsing and subsequent processing.

[0111] Specifically, the analysis method is as follows: analyze the feature histogram H and the object category feature S, and quickly query the possible areas where the user is currently located through the feature-->location reverse query system built during the environmental information collection configuration, and obtain m candidate areas;

[0112] The received image feature set F is parsed, and for each candidate region, a corresponding region BoW search is performed to find up to 10 candidate images.

[0113] Let the image feature set F be represented as: feature word set w = {w1, w2, ..., wn};

[0114] From each region BoW, find multiple images that contain any element of the feature word w (the feature set of each image is ki), and calculate the parameters ai and bi; where ai is the logarithm of the number of identical elements in ki and w, and bi is the ratio of ai to the total number of elements in w.

[0115] Sort all the selected images according to bi from largest to smallest;

[0116] Extract no more than 10 corresponding images with bi > 0.35 from largest to smallest as candidate location images; the number of candidate images can be up to 10. If no candidate image is found, the search fails and returns an error message.

[0117] Furthermore, the present invention also includes a client comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method;

[0118] Furthermore, the present invention also includes a computer-readable storage medium storing a computer program and applied to a client, wherein the computer program, when executed by a processor, implements the steps of the above method.

[0119] Example 4

[0120] Figure 6 This is a schematic diagram of the call unit of the system of the present invention. The call unit includes an indicator light 501, a call button 502, a sensor 503, and a watch strap 504. Further, the sensor 503 is divided into an accelerometer 5030 and a gyroscope 5031.

[0121] The indicator light 501 remains constantly lit when the call center is in operation;

[0122] When the call button 502 is pressed by the user, it sends a call alarm signal, including real-time human body positioning and motion status information, to the property management center of the user's area.

[0123] The sensor 503 is used to acquire the user's motion status information;

[0124] The watch strap 504 is used to secure the call unit to the user's wrist.

[0125] During implementation, the sensor 503 continuously collects corresponding environmental data, and the processors built into the accelerometer 5030 and the gyroscope 5031 analyze the environmental data to obtain the user's motion state information. The motion state information includes the user's human body acceleration data and human body angular velocity data.

[0126] Example 5

[0127] Figure 7 This is a schematic diagram of the call unit of the system of the present invention. The call unit includes an indicator light 501, a call button 502, a sensor 503, and a lanyard 504. Further, the sensor 503 is divided into an accelerometer 5030 and a gyroscope 5031.

[0128] The indicator light 501 remains constantly lit when the call center is in operation;

[0129] When the call button 502 is pressed by the user, it sends a call alarm signal, including real-time human body positioning and motion status information, to the property management center of the user's area.

[0130] The sensor 503 is used to acquire the user's motion status information;

[0131] The lanyard 504 is used to hang the call unit around the user's neck.

[0132] During implementation, the sensor 503 continuously collects corresponding environmental data, and the processors built into the accelerometer 5030 and the gyroscope 5031 analyze the environmental data to obtain the user's motion state information. The motion state information includes the user's human body acceleration data and human body angular velocity data.

[0133] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A positioning method for a wearable device, characterized in that, Includes the following steps: Based on the extraction and differentiation of global and local features of objects and histograms in scene images, different regions are divided, and a word band model of local features is used to construct feature image feature indices for the regions; wherein, dividing different regions specifically involves: The user puts on a vest with the device and turns on the image acquisition mode of the vest. At the same time, the user sits in a smart navigation wheelchair and turns on the real-time positioning function to move around the environment and collect scene data. The scene data includes multiple sets of [P,I1,I2], where P represents the location and direction on the electronic map where the data is collected, and I1 and I2 represent the current images captured by the front and rear cameras, respectively. For each image I in multiple sets [P,I1,I2], extract the color histogram sequence H based on chromaticity, the sequence phrase S based on object detection, and the low-level feature group F based on ORB to form the feature sequence {P, [H,S,F]}, which constitutes a mapping from location to image features. Based on the combination of {P, [H,S,F]}, {[P,H,S]} is extracted to form a new combination data. K-means clustering is used to process the new combination data and divide the user's scene into K regional categories. When using K-means clustering to process the new combination data, for two different data points a=[Pa,Ha,Sa] and b=[Pb,Hb,Sb], the core distance is calculated using the following formula: dis(a,b)=|Pa-Pb|+ɑ||Ha-Hb||+β(Sa, Sb); Where, |Pa-Pb|= , representing the distance between the coordinates of two points; ||Ha-Hb||=∑(| - |) represents the sum of the absolute values ​​of the differences between corresponding terms, calculated using the normalized result; (Sa,Sb)=i, i represents the sum of the number of objects of different types in Sa and Sb; ɑ and β are coefficients, respectively; When there is a positioning requirement, the user-worn device is activated to capture scene images, extract the image features and local feature sets, and transmit the image features and local feature sets to the central control system in real time via relay for identification. The identification process is based on object and histogram features to distinguish the candidate regions where the features are located. Using the image feature index of the candidate region, the candidate location and the image of the candidate location are obtained, and then the matching with the candidate image is completed through the local feature set.

2. The method according to claim 1, characterized in that, The extraction of the image features includes: Extract the [H,S] features of the image, and find the m closest region categories among the K categories of the [H,S] features; Extract the ORB features of the image to obtain the positions of n feature points and their corresponding feature values; Multiple feature values ​​are searched in BoW to find their corresponding feature words and form a feature word set; The image is extracted and stored based on the feature word set as a candidate location image; The position transformation matrix between the candidate position images is calculated, and the matching error is calculated. The candidate position image with the smallest matching error is used as the initial positioning point, and the corresponding shooting position is accurately calculated to obtain the real-time positioning of the user's human body.

3. The method according to claim 2, characterized in that, After extracting the [H,S] features and ORB features of the image, the process includes: When both front and rear camera images have the aforementioned [H,S] features and ORB features, discard the image with fewer feature points and retain the image with the most feature points; If the front and rear camera images do not have the aforementioned [H,S] and ORB features, they are discarded.

4. The method according to claim 2, characterized in that, The process of searching multiple feature values ​​in BoW to find their corresponding feature words and forming a feature word set includes: Obtain the feature word set w = {w1, w2, ..., wn}; Set a specified threshold for the number of elements in the feature word set; If n is less than the specified threshold, the currently processed image is invalid, and the process jumps to the next frame for processing.

5. The method according to claim 2, characterized in that, The step of extracting and storing the image as a candidate location image based on the feature word set includes: Obtain the feature word set ki; Extract the corresponding stored images that contain the element w in the feature word set ki, and calculate the parameter bi of the stored images; The keyframes selected are sorted from largest to smallest by the parameter bi, and no more than 10 corresponding stored images are extracted as candidate location images.

6. The method according to claim 5, characterized in that, After extracting the corresponding stored images containing the element w in the feature word set ki, the process includes: Calculate parameter ai, where ai is the logarithm of the stored image corresponding to the w element contained in the feature word set ki.

7. The method according to claim 6, characterized in that, The parameter bi is the ratio of the parameter ai to the total number of w elements in the feature word set ki.

8. The method according to claim 5, characterized in that, The bi parameter value corresponding to the extracted and stored image is greater than 0.

35.

9. A positioning system for a wearable device, characterized in that, The system includes: An image acquisition module is used to activate the front and rear dual cameras of the user-worn device to capture scene images, and transmit the images in real time to the central control system for identification via a wireless relay through a wireless module; wherein, before capturing the scene images, the system also includes: The user puts on a vest with the device and turns on the image acquisition mode of the vest. At the same time, the user sits in a smart navigation wheelchair and turns on the real-time positioning function to move around the environment and collect scene data. The scene data includes multiple sets of [P,I1,I2], where P represents the location and direction on the electronic map where the data is collected, and I1 and I2 represent the current images captured by the front and rear cameras, respectively. For each image I in multiple sets [P,I1,I2], extract the color histogram sequence H based on chromaticity, the sequence phrase S based on object detection, and the low-level feature group F based on ORB to form the feature sequence {P, [H,S,F]}, which constitutes a mapping from location to image features. Based on the combination of {P, [H,S,F]}, {[P,H,S]} is extracted to form a new combination data. K-means clustering is used to process the new combination data and divide the user's scene into K regional categories. When using K-means clustering to process the new combination data, for two different data points a=[Pa,Ha,Sa] and b=[Pb,Hb,Sb], the core distance is calculated using the following formula: dis(a,b)=|Pa-Pb|+ɑ||Ha-Hb||+β(Sa, Sb); Where, |Pa-Pb|= , representing the distance between the coordinates of two points; ||Ha-Hb||=∑(| - |) represents the sum of the absolute values ​​of the differences between corresponding terms, calculated using the normalized result; (Sa,Sb)=i, i represents the sum of the number of objects of different types in Sa and Sb; ɑ and β are coefficients, respectively; The feature extraction module extracts the image features, obtains the positions of n feature points and their corresponding feature values, and searches for the multiple feature values ​​in BoW to find their corresponding feature words and form a feature word set.

10. The system according to claim 9, characterized in that, The system also includes: A candidate location image generation module, wherein the candidate location image generation module extracts and stores the image based on the feature word set as a candidate location image; The real-time human body positioning module calculates the position transformation matrix between the candidate position images and calculates the matching error. It uses the candidate position image with the smallest matching error as the initial positioning point and accurately calculates the corresponding shooting position to obtain the user's real-time human body positioning.

11. The system according to claim 10, characterized in that, The system also includes a call unit, which automatically acquires the real-time human location calculated by the real-time human location module, and sends a call alarm signal containing the real-time human location of the user to the property management center of the user's area when it receives a call command.

12. A client comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 8.

13. A computer-readable storage medium storing a computer program and applicable to a client, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images