A target scene recognition method and system based on AR glasses

Abstract

The present application relates to the technical field of AR devices, and discloses a target scene recognition method and system based on AR glasses. The present invention sets a recognition interval, and compares the frame diagram of a person to be recognized within the recognition interval by the difference between the light angle and the optimal shooting light angle, thereby improving the recognition accuracy of the person in the target scene diagram affected by the light angle. Compared with the traditional method of setting a recognition threshold, the present invention can perform secondary recognition in the frame diagram of a person to be recognized whose similarity is close to the recognition threshold, thereby effectively improving the recognition efficiency of the frame diagram of a person to be recognized whose similarity is close to the recognition threshold.

Description

Technical Field

[0001] The present application relates to the technical field of AR devices, and in particular to a target scene recognition method and system based on AR glasses. Background Art

[0002] Although person identification has been widely used, many issues still exist in the process. For example, the accuracy of person identification can be affected by lighting conditions. While this issue can lead to a poor user experience in daily use, it generally has minimal negative consequences. However, some AR devices, such as AR glasses, are already integrated with public security systems. If this problem affects the accuracy of identification during patrols or security personnel's patrols, it can significantly hinder patrols and law enforcement. Currently, there is a lack of methods and systems for person identification in the target scenarios of AR glasses that are compatible with AR glasses.

[0003] In view of this, the present invention proposes a target scene recognition method and system based on AR glasses, which are used to eliminate light interference in the target scene to greatly improve the recognition accuracy, and assist in the application of AR glasses and security patrol processes. Summary of the Invention

[0004] In order to eliminate light interference in the target scene to greatly improve the recognition accuracy and assist in the application of AR glasses and security patrol processes, the present application provides a target scene recognition method and system based on AR glasses.

[0005] In a first aspect, the present application provides a method and system for target scene recognition based on AR glasses, which adopts the following technical solutions:

[0006] A target scene recognition method based on AR glasses includes the following steps:

[0007] Obtain a target scene image and extract a frame image of a person to be identified from the target scene image;

[0008] Extract feature vectors based on the image of the person to be identified and compare them with feature vectors in a pre-stored database to calculate the similarity between the two feature vectors. Before obtaining the feature vectors, the collected image needs to be preprocessed to improve image quality, enhance the person's features in the image, and reduce the impact of noise on subsequent processing. Common preprocessing operations include grayscale conversion, noise reduction, and normalization.

[0009] Set an identification interval and compare the similarity with the identification interval. If the similarity is greater than the value of the right endpoint of the identification interval, the comparison result is output as a completed match. If the similarity falls within the identification interval, the comparison result is output as to be analyzed. If the similarity is less than the value of the left endpoint of the identification interval, the comparison result is output as unmatched. It should be noted that the feature vectors describing different character information all correspond to recognition thresholds, and the corresponding recognition thresholds are included in the identification interval. Different recognition intervals are set for the feature vectors describing different character information. In addition, if there are multiple characters to be identified whose similarities are greater than the value of the right endpoint of the identification interval during the matching process, the one with the largest similarity is selected as the matching result of the completed match.

[0010] For the frame image of the person to be identified, the preset optimal shooting light angle and angle threshold are obtained, and then the shooting light angle is output based on the target scene image. The output shooting light angle can be performed by brightness, contrast, or using a light analysis plug-in. The specific method is not limited, and the method of using image brightness for analysis is preferred. In the present invention, the light angle is calculated by the angle between the light and the ground;

[0011] The absolute value of the difference between the shooting light angle and the optimal shooting light angle is recorded as the angle difference, and the angle difference is compared with the angle threshold;

[0012] If the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the device wearer to move along the guide arrow and continuously obtain the frame image of the person to be identified during the movement along the guide arrow. The frame image of the person to be identified is used to perform secondary recognition on the frame image of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame image of the person to be identified cannot be matched.

[0013] Through the above technical solution: a technical solution for identifying people in a target scene is provided. The present invention sets an identification interval, and compares the frame diagram of the person to be identified within the identification interval by the difference between the light angle and the optimal shooting light angle, thereby improving the recognition accuracy of the person in the target scene diagram affected by the light angle. Compared with the traditional method of setting the recognition threshold, secondary recognition can be performed in the frame diagram of the person to be identified whose similarity is close to the recognition threshold, thereby effectively improving the recognition efficiency of the frame diagram of the person to be identified whose similarity is close to the recognition threshold.

[0014] Optionally, the feature vector is used to describe one of appearance information, posture information and clothing information.

[0015] Optionally, the process of setting the identification interval includes:

[0016] Obtaining feature vectors and corresponding description information, and obtaining recognition thresholds and corresponding recognition accuracy distribution data based on historical data of the corresponding description information;

[0017] Set a benchmark recognition accuracy rate and an accuracy rate increase, obtain the recognition threshold at the benchmark recognition rate through distribution data as the first threshold, and the recognition threshold corresponding to the sum of the benchmark recognition accuracy rate and the accuracy rate increase as the second threshold. A successful match does not mean that the recognition is correct. The accuracy rate of the present invention is the ratio of the correct recognition to the total number of recognitions under the successful matching condition;

[0018] The midpoint of the identification interval is the first threshold, and the length of the identification interval is twice the absolute value of the difference between the first threshold and the second threshold.

[0019] Through the above technical solution: a process of setting a distinction interval is given. The distinction interval of the present invention is based on the benchmark recognition accuracy rate, and further judgment analysis can be provided within the range of accuracy increase above and below the benchmark recognition accuracy rate. On the one hand, the problem of recognition errors can be eliminated as much as possible, and on the other hand, some people who were originally judged to be unrecognizable can be accurately identified, thereby effectively improving the recognition accuracy rate.

[0020] Optionally, the process of generating a guide arrow to guide the device wearer to move along the guide arrow includes:

[0021] The direction of the guide arrow is from the current shooting position to the direction in which the absolute value of the difference is reduced;

[0022] The guide arrow also includes a restriction frame, and the device wearer maintains the person to be identified within the restriction frame while moving along the direction of the guide arrow. That is, the device wearer needs to turn his head while moving to keep the person to be identified within the restriction frame during the movement.

[0023] Optionally, the process of performing secondary recognition on the frame of the person to be recognized includes:

[0024] Recording the target scene image during movement along the direction of the guide arrow as a frame image, obtaining the similarity between the frame image of the person to be identified and the current object to be analyzed, and generating a curve of the change of the similarity with the shooting light angle, wherein the object to be analyzed is a feature vector in a pre-stored database whose similarity with the feature vector extracted from the frame image of the person to be identified falls within the recognition interval;

[0025] A correction amount is obtained based on the change curve, and then the similarity between the frame image of the person to be identified and the current object to be analyzed is added to the correction amount to obtain a correction value, and the correction value is compared with the first threshold. If the correction value is greater than the first threshold, the secondary recognition result is output as a match between the frame image of the person to be identified and the current object to be analyzed; otherwise, the secondary recognition result is output as a mismatch.

[0026] Optionally, the process of obtaining the correction amount based on the change curve includes:

[0027] Get the change curve extreme points and arrange them in the order of acquisition, using the formula Get the correction factor ,in is the shooting light angle of the frame image corresponding to the i-th extreme point after arrangement, The similarity between the frame image of the person to be identified and the current object to be analyzed is generated and a curve of similarity changes with the shooting light angle is generated, i is not greater than A non-zero positive integer, is the default selection degree;

[0028] Substitute the correction coefficient into the preset conversion function to calculate the correction amount.

[0029] Through the above technical solution: the process of obtaining the correction coefficient is improved. The correction coefficient selects the extreme point where the similarity change direction of the change curve changes the most as the calculation position, which is used to reflect the influence of light on the similarity in the current target scene image. Since the influence of light on similarity is nonlinear and different eigenvectors are affected by light in different states, and the direction of similarity change at the extreme point changes, it is the position with the most drastic change direction. That is, the relationship between the correction coefficient obtained through the data of the extreme point and the influence of light on similarity is the most reliable.

[0030] Optionally, the preset conversion function is obtained by:

[0031] By formula ; The correction amount is , The basic angle value is the average of the sum of all differences in the shooting light angles of adjacent extreme points after arrangement. The preset conversion function sets different magnifications under different correction coefficients based on empirical data to provide the most reliable data.

[0032] Optionally, if the output result is that the current frame of the person to be identified cannot be matched, a new person number and corresponding feature vector are created for the person in the current frame of the person to be identified and then uploaded and stored in a pre-stored database.

[0033] Optionally, based on the matched frame diagram of the person to be identified, information of the matching person in the database is obtained and the device wearer is notified after the preset requirements are met. The method for determining whether the preset requirements are met is text content analysis.

[0034] In a second aspect, the present application provides a target scene recognition system based on AR glasses, comprising:

[0035] A data acquisition module, wherein the data acquisition module is used to acquire a target scene image;

[0036] An image analysis module, which extracts a feature vector based on the frame image of the person to be identified and compares the feature vector with a feature vector pre-stored in a database and calculates the similarity between the two feature vectors;

[0037] An identification and judgment module is provided with an identification interval, which compares the similarity with the identification interval. If the similarity is greater than the value of the right endpoint of the identification interval, the comparison result is output as a completed match. If the similarity falls within the identification interval, the comparison result is output as to be analyzed. If the similarity is less than the value of the left endpoint of the identification interval, the comparison result is output as unmatched. The identification and judgment module obtains a preset optimal shooting light angle and an angle threshold for the frame image of the person to be identified to be analyzed, and then outputs the shooting light angle based on the target scene image; the absolute value of the difference between the shooting light angle and the optimal shooting light angle is recorded as the angle difference, and the angle difference is compared with the angle threshold; if the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the wearer of the device to move along the guide arrow and the frame image of the person to be identified is continuously obtained during the movement along the guide arrow, and the frame image of the person to be identified is secondary recognized by the frame image of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame image of the person to be identified cannot be matched;

[0038] The notification module obtains the information of the matching person in the database based on the frame diagram of the person to be identified that has been matched, and notifies the device wearer after the preset requirements are met. The method for determining whether the preset requirements are met is text content analysis.

[0039] In summary, this application includes at least one of the following beneficial technical effects:

[0040] The present invention sets an identification interval, and compares the frame diagram of the person to be identified within the identification interval by the difference between the light angle and the optimal shooting light angle, thereby improving the recognition accuracy of the person in the target scene diagram affected by the light angle. Compared with the traditional method of setting the recognition threshold, secondary recognition can be performed in the frame diagram of the person to be identified whose similarity is close to the recognition threshold, thereby effectively improving the recognition efficiency of the frame diagram of the person to be identified whose similarity is close to the recognition threshold.

[0041] The distinction interval of the present invention is based on the benchmark recognition accuracy rate, and can provide further judgment analysis within the range of accuracy increase above and below the benchmark recognition accuracy rate. On the one hand, it can eliminate the problem of recognition errors as much as possible, and on the other hand, it can accurately identify some people who were originally judged to be unrecognizable, thereby effectively improving the recognition accuracy rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the steps of the identification method of the present invention.

[0043] Figure 2 It is a schematic diagram of the process steps for setting the identification interval of the present invention. DETAILED DESCRIPTION

[0044] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0045] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with an embodiment or example is included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0046] The present application embodiment discloses a target scene recognition method based on AR glasses, referring to Figure 1 , including the following steps:

[0047] S100, obtaining a target scene image and extracting a frame image of a person to be identified from the target scene image;

[0048] S200, extracting a feature vector based on the frame image of the person to be identified and comparing the feature vector with the feature vector in a pre-stored database and calculating the similarity between the two feature vectors. Before obtaining the feature vector, the collected image needs to be preprocessed to improve image quality, enhance the person features in the image, and reduce the impact of noise on subsequent processing. Common preprocessing operations include grayscale conversion, noise reduction, normalization, etc.

[0049] S300, setting an identification interval, and comparing the similarity with the identification interval to obtain a comparison result, wherein if the similarity is greater than the value of the right endpoint of the identification interval, the comparison result is output as a completed match; if the similarity falls within the identification interval, the comparison result is output as a pending analysis; if the similarity is less than the value of the left endpoint of the identification interval, the comparison result is output as an unmatched match. It should be noted that each feature vector describing different person information corresponds to a recognition threshold, and the corresponding recognition threshold is included in the identification interval. Different recognition intervals are set for feature vectors describing different person information. In addition, if there are multiple person images to be identified whose similarities are greater than the value of the right endpoint of the identification interval during the matching process, the one with the greatest similarity is selected as the matching result for the completed match;

[0050] S400: For the frame image of the person to be identified, a preset optimal shooting light angle and an angle threshold are obtained, and then the shooting light angle is output based on the target scene image. The output shooting light angle can be output by brightness, contrast, or using a light analysis plug-in. The specific method is not limited, but preferably the method of analyzing the image brightness is used. In the present invention, the light angle is calculated by the angle between the light and the ground.

[0051] S500, recording the absolute value of the difference between the shooting light angle and the optimal shooting light angle as the angle difference, and comparing the angle difference with the angle threshold;

[0052] S600. If the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the device wearer to move along the guide arrow and the frame image of the person to be identified is continuously obtained during the movement along the guide arrow. The frame image of the person to be identified is used to perform secondary recognition on the frame image of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame image of the person to be identified cannot be matched.

[0053] In this embodiment, a technical solution for identifying people in a target scene is provided. An identification interval is set in the present invention. The frame diagram of the person to be identified within the identification interval is compared by the difference between the light angle and the optimal shooting light angle, thereby improving the recognition accuracy of the person in the target scene diagram affected by the light angle. Compared with the traditional method of setting the recognition threshold, secondary recognition can be performed in the frame diagram of the person to be identified whose similarity is close to the recognition threshold, thereby effectively improving the recognition efficiency of the frame diagram of the person to be identified whose similarity is close to the recognition threshold.

[0054] The feature vector is used to describe one of appearance information, posture information and clothing information.

[0055] Reference Figure 2 ,The process of setting the recognition interval includes:

[0056] S310, obtaining a feature vector and corresponding description information, and obtaining a recognition threshold and corresponding recognition accuracy distribution data based on historical data of the corresponding description information;

[0057] S320: Set a benchmark recognition accuracy rate and an accuracy rate increase. Obtain a recognition threshold at the benchmark recognition rate using distribution data as a first threshold. A recognition threshold corresponding to the sum of the benchmark recognition accuracy rate and the accuracy rate increase is a second threshold. A successful match does not necessarily mean correct recognition. The accuracy rate of the present invention is the ratio of correct recognition to the total number of recognitions under successful matching conditions.

[0058] S330 , the midpoint of the identification interval is the first threshold, and the length of the identification interval is twice the absolute value of the difference between the first threshold and the second threshold.

[0059] In this embodiment, the process of setting the distinction interval is given. The distinction interval of the present invention is based on the benchmark recognition accuracy rate, and further judgment analysis can be provided within the range of accuracy increase above and below the benchmark recognition accuracy rate. On the one hand, the problem of recognition errors can be eliminated as much as possible, and on the other hand, some people who were originally judged to be unrecognizable can be accurately identified, thereby effectively improving the recognition accuracy rate.

[0060] The process of generating a guide arrow to guide the device wearer to move along the guide arrow includes:

[0061] S610: The direction of the guide arrow is from the current shooting position to the direction that reduces the absolute value of the difference;

[0062] S620. The guide arrow also includes a restriction frame. The device wearer maintains the person to be identified within the restriction frame while moving along the direction of the guide arrow. That is, the device wearer needs to move while turning his head to keep the person to be identified within the restriction frame during the movement.

[0063] S630, the process of performing secondary recognition on the frame of the person to be recognized includes:

[0064] S640, recording the target scene image during the movement along the direction of the guide arrow as a frame image, obtaining the similarity between the frame image of the person to be identified and the current object to be analyzed, and generating a curve of the change of the similarity with the shooting light angle, wherein the object to be analyzed is a feature vector in a pre-stored database whose similarity with the feature vector extracted from the frame image of the person to be identified falls within the recognition interval;

[0065] S650. Obtain a correction amount based on the change curve, then add the similarity between the frame image of the person to be identified and the current object to be analyzed and the correction amount to obtain a correction value, and compare the correction value with the first threshold value. If the correction value is greater than the first threshold value, output a secondary recognition result that the frame image of the person to be identified in the frame image matches the current object to be analyzed; otherwise, output a secondary recognition result that does not match.

[0066] The process of obtaining the correction value based on the change curve includes:

[0067] Get the change curve extreme points and arrange them in the order of acquisition, using the formula Get the correction factor ,in is the shooting light angle of the frame image corresponding to the i-th extreme point after arrangement, The similarity between the frame image of the person to be identified and the current object to be analyzed is generated and a curve of similarity changes with the shooting light angle is generated, i is not greater than A non-zero positive integer, It is the preset selection degree, which is an angle value, and the value is selected between 0.1-0.3, preferably 0.1;

[0068] Substitute the correction coefficient into the preset conversion function to calculate the correction amount.

[0069] In this embodiment, the process of obtaining the correction coefficient is improved. The correction coefficient selects the extreme point where the similarity change direction of the change curve changes the most as the calculation position to reflect the influence of light on the similarity in the current target scene image. Since the influence of light on the similarity is nonlinear and different eigenvectors are affected by light in different states, and the direction of similarity change at the extreme point changes, it is the position where the change direction is most drastic. That is, the relationship between the correction coefficient obtained through the data of the extreme point and the influence of light on the similarity is the most reliable.

[0070] The methods for obtaining the preset conversion function include:

[0071] By formula ; The correction amount is , The basic angle value is the average of the sum of all differences in the shooting light angles of adjacent extreme points after arrangement. The preset conversion function sets different magnifications under different correction coefficients based on empirical data to provide the most reliable data.

[0072] If the output result is that the current person frame to be identified cannot be matched, a new person number and corresponding feature vector are created for the person in the current person frame to be identified and then uploaded and stored in the pre-stored database.

[0073] Based on the matched frame diagram of the person to be identified, the information of the matching person in the database is obtained and the device wearer is notified after the preset requirements are met. The judgment method for meeting the preset requirements is text content analysis.

[0074] In a second aspect, the present application provides a target scene recognition system based on AR glasses, comprising:

[0075] Data acquisition module, the data acquisition module is used to obtain target scene images;

[0076] Image analysis module, which extracts feature vectors based on the frame image of the person to be identified and compares the feature vectors with feature vectors pre-stored in the database and calculates the similarity between the two feature vectors;

[0077] The recognition and judgment module is provided with a recognition interval, which compares the similarity with the recognition interval. If the similarity is greater than the value of the right endpoint of the recognition interval, the comparison result is output as a completed match. If the similarity falls within the recognition interval, the comparison result is output as to be analyzed. If the similarity is less than the value of the left endpoint of the recognition interval, the comparison result is output as unmatched. The recognition and judgment module obtains the preset optimal shooting light angle and angle threshold for the frame diagram of the person to be identified to be analyzed, and then outputs the shooting light angle based on the target scene image; the absolute value of the difference between the shooting light angle and the optimal shooting light angle is recorded as the angle difference, and the angle difference is compared with the angle threshold; if the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the wearer of the device to move along the guide arrow and the frame diagram of the person to be identified is continuously obtained during the movement along the guide arrow, and the frame diagram of the person to be identified is secondary recognized by the frame diagram of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame diagram of the person to be identified cannot be matched;

[0078] The notification module obtains the information of the matching person in the database based on the frame diagram of the person to be identified that has been matched, and notifies the device wearer after the preset requirements are met. The judgment method for meeting the preset requirements is text content analysis.

[0079] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A target scene recognition method based on AR glasses, characterized in that: The steps include: Obtain a target scene image and extract a frame image of a person to be identified from the target scene image; Extracting a feature vector based on the frame image of the person to be identified and calculating the similarity between the feature vector and the feature vectors stored in a database; Set the recognition interval and compare the similarity with the recognition interval. If the similarity is greater than the value at the right endpoint of the recognition interval, the comparison result is output as a complete match. If the similarity falls within the recognition interval, the comparison result is output as pending analysis. If the similarity is less than the value at the left endpoint of the recognition interval, the comparison result is output as unmatched. The process of setting the recognition interval includes: Obtaining feature vectors and corresponding description information, and obtaining recognition thresholds and corresponding recognition accuracy distribution data based on historical data of the corresponding description information; Set a benchmark recognition accuracy rate and an accuracy rate increase, obtain the recognition threshold at the benchmark recognition rate through distribution data as a first threshold, and the recognition threshold corresponding to the sum of the benchmark recognition accuracy rate and the accuracy rate increase as a second threshold; The midpoint of the identification interval is the first threshold, and the length of the identification interval is twice the absolute value of the difference between the first threshold and the second threshold; For the frame image of the person to be identified, obtain the preset optimal shooting light angle and angle threshold, and then output the shooting light angle based on the target scene image; The absolute value of the difference between the shooting light angle and the optimal shooting light angle is recorded as the angle difference, and the angle difference is compared with the angle threshold; If the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the device wearer to move along the guide arrow and the frame image of the person to be identified is continuously acquired during the movement along the guide arrow. The frame image of the person to be identified is used for secondary recognition of the frame image of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame image of the person to be identified cannot be matched. The process of secondary recognition of the frame of the person to be recognized includes: Recording the target scene image during movement along the direction of the guide arrow as a frame image, obtaining the similarity between the frame image of the person to be identified and the current object to be analyzed, and generating a curve of the change of the similarity with the shooting light angle, wherein the object to be analyzed is an object whose feature vector in a pre-stored database falls within the recognition interval with the feature vector extracted from the frame image of the person to be identified; A correction amount is obtained based on the change curve, and then the similarity between the frame image of the person to be identified and the current object to be analyzed is added to the correction amount to obtain a correction value, and the correction value is compared with the first threshold. If the correction value is greater than the first threshold, the secondary recognition result is output as a match between the frame image of the person to be identified and the current object to be analyzed; otherwise, the secondary recognition result is output as a mismatch.

2. The target scene recognition method based on AR glasses according to claim 1, characterized in that: The feature vector is used to describe one of appearance information, posture information and clothing information.

3. The target scene recognition method based on AR glasses according to claim 1, characterized in that: The process of generating a guide arrow to guide the device wearer to move along the guide arrow includes: The direction of the guide arrow is from the current shooting position to the direction in which the absolute value of the difference is reduced; The guide arrow further includes a limit frame, and the device wearer maintains the person to be identified within the limit frame when moving along the direction of the guide arrow.

4. The target scene recognition method based on AR glasses according to claim 1, characterized in that: The process of obtaining the correction value based on the change curve includes: Get the change curve extreme points and arrange them in the order of acquisition, using the formula Get the correction factor ,in is the shooting light angle of the frame image corresponding to the i-th extreme point after arrangement, The similarity between the frame image of the person to be identified and the current object to be analyzed is generated and a curve of similarity changes with the shooting light angle is generated, i is not greater than A non-zero positive integer, is the default selection degree; Substitute the correction coefficient into the preset conversion function to calculate the correction amount.

5. The target scene recognition method based on AR glasses according to claim 4 is characterized in that: The methods for obtaining the preset conversion function include: ; The correction amount is , The basic angle value is the average of the sum of all differences in the shooting light angles of adjacent extreme points after arrangement.

6. The target scene recognition method based on AR glasses according to claim 1, characterized in that: If the output result is that the current person frame to be identified cannot be matched, a new person number and corresponding feature vector are created for the person in the current person frame to be identified and then uploaded and stored in the pre-stored database.

7. The target scene recognition method based on AR glasses according to claim 1, characterized in that: Based on the matched person frame to be identified, obtain the information of the matching person in the database and notify the device wearer after meeting the preset requirements.

8. A target scene recognition system based on AR glasses, characterized in that: The target scene recognition method based on AR glasses as claimed in any one of claims 1 to 7 comprises: A data acquisition module, wherein the data acquisition module is used to acquire a target scene image; An image analysis module, which extracts a feature vector based on the frame image of the person to be identified and calculates the similarity between the two feature vectors based on the feature vector and the feature vectors pre-stored in the database; An identification and judgment module is provided with an identification interval, which compares the similarity with the identification interval. If the similarity is greater than the value of the right endpoint of the identification interval, the comparison result is output as a completed match. If the similarity falls within the identification interval, the comparison result is output as to be analyzed. If the similarity is less than the value of the left endpoint of the identification interval, the comparison result is output as unmatched. The identification and judgment module obtains a preset optimal shooting light angle and an angle threshold for the frame image of the person to be identified to be analyzed, and then outputs the shooting light angle based on the target scene image; the absolute value of the difference between the shooting light angle and the optimal shooting light angle is recorded as the angle difference, and the angle difference is compared with the angle threshold; if the angle difference is greater than the angle threshold, a guide arrow is generated based on the difference to guide the wearer of the device to move along the guide arrow and the frame image of the person to be identified is continuously obtained during the movement along the guide arrow, and the frame image of the person to be identified is secondary recognized by the frame image of the person to be identified during the movement along the guide arrow. If the angle difference is not greater than the angle threshold, it is output that the current frame image of the person to be identified cannot be matched; The notification module obtains the information of the matching person in the database based on the frame diagram of the person to be identified that has been matched, and notifies the device wearer after the preset requirements are met.

Images