Image feature processing method, image feature processing device, medium and electronic equipment

Abstract

The present invention provides an image feature processing method, an image feature processing device, a medium, and an electronic device. The image feature processing method includes: processing images collected by a first front-end device to obtain multiple first images containing pedestrians; encrypting each of the first images to obtain an encrypted image; sending a feature extraction request to at least one second front-end device so that the second front-end device and the first front-end device jointly process the encrypted image to extract first image features; and sending a feature search request to the second front-end device so that the second front-end device and the first front-end device jointly obtain matching features of the first image features. The image feature processing method can improve the utilization of computing resources.

Description

Technical Field

[0001] The present invention relates to the field of pedestrian re-identification, and in particular to an image feature processing method, an image feature processing device, a medium and an electronic device. Background Art

[0002] In recent years, person re-identification algorithms have been widely used in offline retail stores and shopping malls, improving the consumer shopping experience by analyzing data such as the number of people, number of visits, and store associations. Person re-identification algorithms involve image feature processing, which wastes computing resources because it does not utilize the available computing power of front-end devices. Instead, it is centralized in high-performance servers, resulting in additional server costs. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the object of the present invention is to provide an image feature processing method, an image feature processing device, a medium and an electronic device, so as to solve the problem of waste of computing resources in the image feature processing method in the prior art.

[0004] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present invention provides an image feature processing method, which is applied to a first front-end device. The image feature processing method includes: processing the images collected by the first front-end device to obtain multiple first images containing pedestrians; encrypting each of the first images to obtain an encrypted image; sending a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract the first image features; sending a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain matching features of the first image features.

[0005] In an embodiment of the first aspect, the method for processing the image captured by the first front-end device includes: detecting and processing the image captured by the first front-end device to obtain a second image containing a pedestrian and pedestrian detection information of the second image; based on the pedestrian detection information, processing the second image to obtain a pedestrian trajectory of the second image; and obtaining the first image based on the pedestrian trajectory.

[0006] In an embodiment of the first aspect, the implementation method of sending a feature extraction request to at least one second front-end device includes: storing the encrypted image in the queue to be extracted of the first front-end device; sending the feature extraction request to the second front-end device in response to a load balancing instruction sent by the load balancing server, the feature extraction request carrying part of the encrypted image in the queue to be extracted.

[0007] In an embodiment of the first aspect, the method for generating the load balancing instruction includes: the load balancing server monitors the first front-end device and the second front-end device to obtain the load information of the first front-end device and the load information of the second front-end device; based on the load information of the first front-end device and the load information of the second front-end device, the load balancing server sends the load balancing instruction to the first front-end device.

[0008] In an embodiment of the first aspect, the image feature processing method further includes: sending matching information and the pedestrian trajectory to the load balancing server, wherein the matching information and the pedestrian trajectory are used to generate a pedestrian trajectory map.

[0009] In an embodiment of the first aspect, the image feature processing method further includes: sending a request instruction to a feature processing server so that the feature processing server receives part of the first image feature, and causing the feature processing server to search its feature library to obtain matching features of the first image feature it received.

[0010] In an embodiment of the first aspect, the method for implementing feature encryption processing on the first image includes: performing feature encryption processing on the first image through a pedestrian image encryption network to obtain the encrypted image, the pedestrian image encryption network is the head of the pedestrian feature extraction network, and the pedestrian feature extraction network is used by the first front-end device and the second front-end device to extract the first image features.

[0011] The second aspect of the present invention provides an image feature processing device, which is applied to a first front-end device, including: a first image acquisition module, used to process the images collected by the first front-end device to obtain multiple first images containing pedestrians; an encryption processing module, used to encrypt each of the first images to obtain an encrypted image; a feature extraction request sending module, used to send a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract the first image features; a feature search request sending module, used to send a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain the matching features of the first image features.

[0012] A third aspect of the present invention provides a computer-readable storage medium, which, when executed by a processor, implements the image feature processing method described in any one of the first aspects of the present invention.

[0013] The fourth aspect of the present invention provides an electronic device, comprising: a memory storing a computer program; a processor communicatively connected to the memory, for executing any one of the image feature processing methods described in the first aspect of the present invention when the computer program is called; and a display communicatively connected to the processor and the memory, for displaying a GUI interactive interface related to the image feature processing method.

[0014] As described above, the image feature processing method, image feature processing device, medium, and electronic device of the present invention have the following beneficial effects:

[0015] The image feature processing method includes processing the images collected by the first front-end device to obtain multiple first images containing pedestrians; encrypting each of the first images to obtain an encrypted image; sending a feature extraction request to at least one second front-end device so that the second front-end device and the first front-end device jointly process the encrypted image to extract the first image features; and sending a feature search request to the second front-end device so that the second front-end device and the first front-end device jointly obtain the matching features of the first image features. By the first front-end device and the second front-end device jointly processing the encrypted image and jointly obtaining the matching features of the first image features, the computing power resources of the first front-end device and the second front-end device can be effectively utilized, thereby avoiding the process of centrally executing the first image processing process only through a high-performance server or only through the first front-end device. Therefore, the image feature processing method can improve the utilization rate of computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Shown is a flowchart of the image feature processing method described in an embodiment of the present invention.

[0017] Figure 2 Shown is a flowchart of an implementation method for processing images captured by the first front-end device in an embodiment of the present invention.

[0018] Figure 3 Shown is a flowchart of an implementation method for sending a feature extraction request to at least one second front-end device in an embodiment of the present invention.

[0019] Figure 4 Shown is a flow chart of a method for generating a load balancing instruction according to an embodiment of the present invention.

[0020] Figure 5 Shown is a structural schematic diagram of the image feature processing device according to an embodiment of the present invention.

[0021] Figure 6 Shown is a schematic structural diagram of the electronic device according to an embodiment of the present invention.

[0022] Component number description

[0023] 500 Image Feature Processing Device

[0024] 510 First Image Acquisition Module

[0025] 520 encryption processing module

[0026] 530 Feature extraction request sending module

[0027] 540 Feature search request sending module

[0028] 600 Electronic Equipment

[0029] 610 Memory

[0030] 620 processor

[0031] 630 Display

[0032] Steps S11-S14

[0033] Steps S21-S23

[0034] Steps S31-S32

[0035] Steps S41-S42 DETAILED DESCRIPTION

[0036] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0037] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0038] In current person re-identification algorithms, image feature processing does not utilize the spare computing resources of the front-end device, but is instead centralized in high-performance servers, resulting in additional server costs. Therefore, current image feature processing methods suffer from the problem of wasted computing resources. To address at least the aforementioned issues, the present invention provides an image feature processing method, applied to a first front-end device, comprising: processing images captured by the first front-end device to obtain multiple first images containing pedestrians; encrypting each of the first images to obtain encrypted images; sending a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted images to extract first image features; and sending a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain matching features of the first image features. By having the first and second front-end devices jointly process the encrypted images and jointly obtain matching features of the first image features, the computing resources of the first and second front-end devices can be effectively utilized, thereby avoiding the need to centrally execute the first image processing process solely on a high-performance server or solely on the first front-end device. Therefore, the image feature processing method can improve the utilization of computing resources.

[0039] In one embodiment of the present invention, the image feature processing method is applied to a first front-end device, see Figure 1 , the image feature processing method includes:

[0040] S11: Process the images collected by the first front-end device to obtain a plurality of first images containing pedestrians. The images collected by the first front-end device may be in the form of surveillance videos.

[0041] Optionally, the implementation method of processing the image collected by the first front-end device may include: processing the image collected by the first front-end device to obtain a pedestrian trajectory; based on the pedestrian trajectory, obtaining an image of a pedestrian on the pedestrian trajectory, the pedestrian image being the first image.

[0042] S12: Encrypt each of the first images to obtain an encrypted image, wherein the encrypted image may be in the form of floating-point features.

[0043] Optionally, the method for encrypting each first image includes: encrypting the first image using a pedestrian image encryption network to obtain the encrypted image, wherein the pedestrian image encryption network is the head of a pedestrian feature extraction network and is a lightweight neural network, and the pedestrian feature extraction network is used by the first front-end device and the second front-end device to extract the first image features in step S13. For example, when the pedestrian feature extraction network is a ResNet network, the structure of the pedestrian image encryption network is:

[0044] Conv2d(kennel_size=7×7, ch_in=3, ch_out=64, stride=2, padding=3) BatchNorm2d(ch_in=64,ch_out=64) ReLU MaxPool(kernel_size=3, stride=2, padding=1)

[0045] Where Conv2d is a convolutional layer, BatchNorm2d is a batch normalization layer, ReLU is an activation function, MaxPool is a maximum pooling layer, kennel size is the kernel size, ch in is the number of input channels, ch out is the number of output channels, stride is the stride, and padding is the padding size. By encrypting the first image, it is possible to ensure that the first image does not leak the privacy of pedestrians during transmission, thereby improving security.

[0046] S13: Send a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract the first image feature. The hardware configuration of the second front-end device can be identical to that of the first front-end device.

[0047] Optionally, the implementation method of the second front-end device and the first front-end device jointly processing the encrypted image includes: the second front-end device and the first front-end device process the encrypted image through the feature extraction network to obtain the first image feature, and the first image feature can be a pedestrian feature in the first image.

[0048] Optionally, another implementation method for the second front-end device and the first front-end device to jointly process the encrypted image includes: if the performance of the first front-end device is sufficient to process all the encrypted images, the first front-end device processes all the encrypted images through the feature extraction network to obtain the first image features.

[0049] Optionally, the method for sending a feature extraction request to at least one second front-end device so that the second front-end device and the first front-end device jointly process the encrypted image includes: the first front-end device sending the feature extraction request to the second front-end device, and the first and second front-end devices performing feature extraction processing on the encrypted image using a person re-identification model to obtain the first image features. The person re-identification model can be implemented based on a convolutional neural network, which can be a ResNet network.

[0050] Optionally, a method for implementing the second front-end device and the first front-end device processing the encrypted image through the feature extraction network may include: adjusting the speed of extracting the first image features based on the operating status of the first front-end device and the second front-end device, so as to prevent the process of extracting the first image features from occupying a large amount of computing power of the first front-end device and the second front-end device, thereby affecting the image detection and image encryption processes of the first front-end device and the second front-end device. The operating status may include central processing unit information and device temperature information, etc.

[0051] S14: Send a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain matching features of the first image features.

[0052] Optionally, the first image feature includes a first feature to be searched and a second feature to be searched. The first feature to be searched may be the image feature to be searched by the first front-end device, and the second feature to be searched may be the image feature to be searched by the second front-end device. A method for the second front-end device and the first front-end device to jointly obtain a matching feature for the first image feature includes: based on the first image feature, the first front-end device and the second front-end device search their respective feature libraries to obtain matching features for the first feature to be searched and the second feature to be searched, respectively. The first feature to be searched is stored in the feature library of the first front-end device, and the second feature to be searched is stored in the feature library of the second front-end device. The feature library may be a database for storing pedestrian features. The matching feature for the first image feature may be the pedestrian feature in the feature library that has the highest similarity to the first image feature, excluding the first image feature. A similarity threshold may also be set, and pedestrian features in the feature library that exceed the similarity threshold are obtained as matching features. This embodiment is not limited to this. The feature library may be a Redis database or a MySQL database.

[0053] Optionally, the implementation method of the second front-end device and the first front-end device jointly obtaining the matching features of the first image features also includes: obtaining matching information based on the matching features of the second feature to be searched; and sending the matching information to the first front-end device. The matching information can be the matching relationship of pedestrian identifications between the front-end devices. For example, if the pedestrian identification of pedestrian A in the image collected by the first front-end device is 35, and the pedestrian identification of pedestrian A in the image collected by the second front-end device is 48, then the pedestrian identification of 35 in the first front-end device matches the pedestrian identification of 48 in the second front-end device. In addition, the pedestrian identification can be used as a trajectory identification of the corresponding pedestrian trajectory. For example, in the first front-end device, the pedestrian identification of pedestrian B is 59, then the trajectory identification of pedestrian B is also 59. By searching for part of the first image feature on the second front-end device, cross-domain matching information can be obtained.

[0054] Optionally, the first front-end device and the second front-end device searching their respective feature libraries also include: the first front-end device and the second front-end device respectively generate a first queue to be searched and a second queue to be searched; respectively store the first feature to be searched in the first queue to be searched and the second feature to be searched in the second queue to be searched; respectively obtain the first feature to be searched and the second feature to be searched from the first queue to be searched and the second queue to be searched.

[0055] According to the above description, the image feature processing method described in this embodiment includes processing the images collected by the first front-end device to obtain multiple first images containing pedestrians; encrypting each of the first images to obtain an encrypted image; sending a feature extraction request to at least one second front-end device so that the second front-end device and the first front-end device jointly process the encrypted image to extract the first image features; and sending a feature search request to the second front-end device so that the second front-end device and the first front-end device jointly obtain the matching features of the first image features. By the first front-end device and the second front-end device jointly processing the encrypted image and jointly obtaining the matching features of the first image features, the computing power resources of the first front-end device and the second front-end device can be effectively utilized, thereby avoiding the process of centrally executing the first image processing process only through a high-performance server or only through the first front-end device. Therefore, the image feature processing method can improve the utilization rate of computing power resources.

[0056] See also Figure 2 In one embodiment of the present invention, a method for processing the image collected by the first front-end device includes:

[0057] S21: Perform detection processing on the image captured by the first front-end device to obtain a second image containing a pedestrian and pedestrian detection information of the second image.

[0058] Optionally, the method for detecting and processing the image captured by the first front-end device may include: processing the image captured by the first front-end device using a target detection algorithm to obtain the second image and pedestrian detection information of the second image, wherein the pedestrian detection information may be a detection box of a pedestrian's torso. The target detection algorithm may be implemented based on a convolutional neural network, which may be a YOLO V4 network or an Efficient Net network.

[0059] S22: Based on the pedestrian detection information, process the second image to obtain pedestrian trajectories in the second image. The pedestrian trajectories in the second image may be a pedestrian trajectory set formed by summarizing each pedestrian trajectory in the second image.

[0060] Optionally, the method for processing the second image may include: processing the second image using a pedestrian tracking algorithm to obtain a pedestrian trajectory in the second image.

[0061] Optionally, a method for processing the second image using a pedestrian tracking algorithm to obtain a pedestrian trajectory in the second image may include: the second image includes a target frame image and a previous frame image of the target frame image; processing the target frame image and the previous frame image using a lightweight neural network to obtain image features of the target frame image and the previous frame image; obtaining feature similarity between the image features of the target frame image and the previous frame image; obtaining spatial similarity between pedestrian detection information of the target frame image and the previous frame image; and processing the target frame image and the previous frame image based on the feature similarity and the spatial similarity to obtain the pedestrian trajectory in the second image. The lightweight neural network may be a MobileNetV2 network or a ShuffleNet network.

[0062] S23: Acquire the first image based on the pedestrian trajectory.

[0063] Optionally, the implementation method of obtaining the first image based on the pedestrian trajectory includes: obtaining images on the pedestrian trajectory with a high confidence level and a certain interval time, for example, obtaining images on the pedestrian trajectory with a confidence level greater than 95% and an interval time greater than 5 seconds, to ensure that the first image can extract pedestrian features, thereby improving image processing efficiency.

[0064] As can be seen from the above description, the method for implementing processing of images captured by the first front-end device in this embodiment includes detecting and processing the images captured by the first front-end device to obtain a second image containing a pedestrian and pedestrian detection information from the second image; processing the second image based on the pedestrian detection information to obtain a pedestrian trajectory in the second image; and obtaining the first image based on the pedestrian trajectory. By processing the images captured by the first front-end device, the pedestrian trajectory and the first image on the pedestrian trajectory can be obtained, thereby achieving human body tracking.

[0065] See also Figure 3 In one embodiment of the present invention, a method for sending a feature extraction request to at least one second front-end device includes:

[0066] S31, storing the encrypted image in a queue to be extracted of the first front-end device.

[0067] Optionally, the implementation method of storing the encrypted image to the queue to be extracted of the first front-end device may include: setting multiple tasks to be processed, where the tasks to be processed may be a collection of the encrypted images, for example, one task to be processed contains ten encrypted images; and storing the tasks to be processed to the queue to be extracted of the first front-end device.

[0068] S32: In response to the load balancing instruction sent by the load balancing server, send the feature extraction request to the second front-end device, where the feature extraction request carries the portion of the encrypted image in the queue to be extracted. The load balancing instruction may carry communication information of the second front-end device, where the communication information of the second front-end device is used for communication between the first and second front-end devices.

[0069] Optionally, the communication protocol between the first front-end device and the second front-end device can be any one of the protobuf protocol, the json protocol and the xml protocol.

[0070] Optionally, the implementation method of sending the feature extraction request to the second front-end device includes: the first front-end device sends the feature extraction request to the second front-end device, and the feature extraction request carries part of the task to be processed; the second front-end device receives the task to be processed.

[0071] As can be seen from the above description, the method for sending a feature extraction request to at least one second front-end device described in this embodiment includes: storing the encrypted image in a queue to be extracted on the first front-end device; and sending the feature extraction request to the second front-end device in response to a load balancing instruction sent by a load balancing server. By sending the feature extraction request to the second front-end device, the encrypted image that has not undergone feature extraction can be sent to the second front-end device, thereby achieving computing power sharing between the first and second front-end devices.

[0072] See also Figure 4 In one embodiment of the present invention, the method for generating the load balancing instruction includes:

[0073] S41: The load balancing server monitors the first front-end device and the second front-end device to obtain load information of the first front-end device and load information of the second front-end device.

[0074] Optionally, the load balancing server is embedded in the first front-end device and the second front-end device to improve the operability of the first front-end device and the second front-end device.

[0075] Optionally, the load information may include processor usage, and a processor usage threshold may be set to determine the load status of the first front-end device and the second front-end device. For example, if the processor usage threshold is 75%, if the processor usage of the first front-end device is greater than 75%, the load status of the first front-end device is a high load status; if the processor usage of the first front-end device is less than 75%, the load status of the first front-end device is a low load status.

[0076] Optionally, the implementation method of the load balancing server monitoring the first front-end device and the second front-end device may include: the load balancing server analyzing the operating status reported by the first front-end device and the second front-end device to obtain the load information of the first front-end device and the load information of the second front-end device.

[0077] S42: Based on the load information of the first front-end device and the load information of the second front-end device, the load balancing server sends the load balancing instruction to the first front-end device.

[0078] Optionally, the first front-end device is in the high-load state, the second front-end device is in the low-load state, and the implementation method of the load balancing server sending the load balancing instruction to the first front-end device includes: the load balancing server obtains the communication information of the second front-end device, and the communication information of the second front-end device is used for communication between the first front-end device and the second front-end device; based on the communication information of the second front-end device, the load balancing server generates the load balancing instruction; the load balancing server sends the load balancing instruction to the first front-end device.

[0079] In one embodiment of the present invention, the image feature processing method further includes: sending a request instruction to a feature processing server so that the feature processing server receives part of the first image feature, and causing the feature processing server to search and process its feature library to obtain matching features of the first image feature it received. The feature processing server may be a load balancing server, and the request instruction carries part of the first image feature. The feature library of the feature processing server may be a database of the feature processing server, and the database is used to store pedestrian features. The matching feature may be the pedestrian feature in the feature library with the highest similarity to the first image feature. A similarity threshold may also be set to obtain pedestrian features in the feature library that are greater than the similarity threshold as matching features. This embodiment is not limited to this.

[0080] As can be seen from the above description, the image feature processing method described in this embodiment includes sending a request instruction to a feature processing server, causing the feature processing server to receive a portion of the first image feature, and then searching its feature library to obtain matching features of the received first image feature. By searching the feature processing server, the image feature processing pressure on the front-end device can be reduced, thereby improving the efficiency of searching for matching features.

[0081] In one embodiment of the present invention, the image feature processing method further includes: sending matching information and the pedestrian trajectory to the load balancing server, wherein the matching information and the pedestrian trajectory are used to generate a pedestrian trajectory map. The matching information can be a matching relationship between pedestrian identifiers between front-end devices.

[0082] Optionally, the method for sending the matching information and the pedestrian trajectory to the load balancing server may include: the first front-end device sending the matching information of the first front-end device and the pedestrian trajectory of the first front-end device to the load balancing server; the second front-end device sending the matching information of the second front-end device and the pedestrian trajectory of the second front-end device to the load balancing server; and based on the matching information of the first front-end device and the matching information of the second front-end device, processing the pedestrian trajectory of the first front-end device and the pedestrian trajectory of the second front-end device to generate the pedestrian trajectory map. The pedestrian trajectory map includes the pedestrian trajectory of each pedestrian under the first front-end device and the second front-end device.

[0083] Optionally, based on the matching information of the first front-end device and the matching information of the second front-end device, the pedestrian trajectory of the first front-end device and the pedestrian trajectory of the second front-end device are processed to generate a pedestrian trajectory map. An implementation method can be: based on the matching information of the first front-end device, the matching information of the second front-end device, the pedestrian trajectory of the first front-end device and the pedestrian trajectory of the second front-end device, the identification file is generated; and the identification file is sent to a cloud server, and the cloud server is used to analyze the identification file to generate the pedestrian trajectory map.

[0084] Optionally, the image feature processing method further includes: sending the number of pedestrians to the load balancing server, where the number of pedestrians is the number of pedestrians in the first image.

[0085] As can be seen from the above description, the image feature processing method described in this embodiment includes sending matching information and the pedestrian trajectory to the load balancing server, where the matching information and the pedestrian trajectory are used to generate a pedestrian trajectory map. By aggregating the matching information and the pedestrian trajectory on the load balancing server, a pedestrian trajectory map can be generated for each pedestrian across all front-end devices, thereby improving pedestrian recognition efficiency.

[0086] In one embodiment of the present invention, an image feature processing device 500 is provided. Specifically, please refer to Figure 5 , the image feature processing device 500 includes:

[0087] A first image acquisition module 510 is configured to process the images captured by the first front-end device to obtain a plurality of first images containing pedestrians;

[0088] An encryption processing module 520, configured to perform encryption processing on each of the first images to obtain an encrypted image;

[0089] a feature extraction request sending module 530, configured to send a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract first image features;

[0090] The feature search request sending module 540 is configured to send a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain matching features of the first image features.

[0091] According to the above description, the image feature processing device processes the encrypted image and obtains the matching features of the first image features jointly through the first front-end device and the second front-end device, which can effectively utilize the computing power resources of the first front-end device and the second front-end device, thereby avoiding the process of centrally executing the first image processing only through a high-performance server or only through the first front-end device. Therefore, the image feature processing method can improve the utilization rate of computing power resources.

[0092] Based on the above description of the label marking method, the present invention also provides a computer readable storage medium on which a computer program is stored. When the computer program is executed by a processor, Figure 1 The image feature processing method shown.

[0093] Based on the above description of the image feature processing method, the present invention also provides an electronic device. Figure 6 In one embodiment of the present invention, the electronic device 600 includes a memory 610 storing a computer program; a processor 620 communicating with the memory 610 and executing the computer program when calling the computer program. Figure 1 The image feature processing method shown; the display 630 is communicatively connected to the processor 620 and the memory 610, and is used to display the relevant GUI interactive interface of the image feature processing method.

[0094] The protection scope of the image feature processing method described in the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing, or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0095] In summary, the image feature processing method, image feature processing device, medium, and electronic device of the present invention are used to improve the utilization rate of computing resources. Therefore, the present invention effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0096] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A method for processing image features, characterized in that: Applied to the first front-end device, the image feature processing method includes: Processing the images captured by the first front-end device to obtain a plurality of first images containing pedestrians; performing encryption processing on each of the first images to obtain an encrypted image; Sending a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract a first image feature, where the first image feature is a feature of a pedestrian in the first image; Sending a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain a matching feature of the first image feature, where the matching feature of the first image feature is a pedestrian feature in a feature library that has the highest similarity to the first image feature, other than the first image feature, or a pedestrian feature in the feature library that has a similarity to the first image feature greater than a similarity threshold; The method for processing the image captured by the first front-end device to obtain multiple first images containing pedestrians includes: performing detection processing on the image captured by the first front-end device to obtain a second image containing a pedestrian and pedestrian detection information of the second image; processing the second image based on the pedestrian detection information to obtain a pedestrian trajectory in the second image; and obtaining the first image based on the pedestrian trajectory; The second front-end device and the first front-end device jointly process the encrypted image to extract first image features, including: adjusting the speed of extracting the first image features based on the operating status of the first front-end device and the second front-end device.

2. The image feature processing method according to claim 1, characterized in that: The implementation method of sending a feature extraction request to at least one second front-end device includes: Storing the encrypted image in a queue to be retrieved on the first front-end device; In response to the load balancing instruction sent by the load balancing server, the feature extraction request is sent to the second front-end device, where the feature extraction request carries part of the encrypted image in the queue to be extracted.

3. The image feature processing method according to claim 2, characterized in that: The method for generating the load balancing instruction includes: The load balancing server monitors the first front-end device and the second front-end device to obtain load information of the first front-end device and load information of the second front-end device; Based on the load information of the first front-end device and the load information of the second front-end device, the load balancing server sends the load balancing instruction to the first front-end device.

4. The image feature processing method according to claim 3, characterized in that: Also includes: The matching information and the pedestrian trajectory are sent to the load balancing server, where the matching information and the pedestrian trajectory are used to generate a pedestrian trajectory map. The matching information is the matching relationship between pedestrian identifiers between front-end devices.

5. The image feature processing method according to claim 1, wherein: Also includes: A request instruction is sent to the feature processing server so that the feature processing server receives part of the first image feature, and the feature processing server searches its feature library to obtain matching features of the received first image feature.

6. The image feature processing method according to claim 1, wherein: The method for encrypting each of the first images to obtain an encrypted image includes: The first image is encrypted through a pedestrian image encryption network to obtain the encrypted image. The pedestrian image encryption network is the head of a pedestrian feature extraction network. The pedestrian feature extraction network is used by the first front-end device and the second front-end device to extract the first image features.

7. An image feature processing device, characterized in that: Applicable to the first front-end device, including: A first image acquisition module, configured to process the images collected by the first front-end device to acquire a plurality of first images containing pedestrians; an encryption processing module, configured to perform encryption processing on each of the first images to obtain an encrypted image; a feature extraction request sending module, configured to send a feature extraction request to at least one second front-end device, so that the second front-end device and the first front-end device jointly process the encrypted image to extract a first image feature, where the first image feature is a feature of a pedestrian in the first image; a feature search request sending module, configured to send a feature search request to the second front-end device, so that the second front-end device and the first front-end device jointly obtain a matching feature of the first image feature, where the matching feature of the first image feature is a pedestrian feature in a feature library that has the highest similarity to the first image feature, other than the first image feature, or a pedestrian feature in the feature library that has a similarity to the first image feature greater than a similarity threshold; The method for processing the image captured by the first front-end device to obtain multiple first images containing pedestrians includes: performing detection processing on the image captured by the first front-end device to obtain a second image containing a pedestrian and pedestrian detection information of the second image; processing the second image based on the pedestrian detection information to obtain a pedestrian trajectory in the second image; and obtaining the first image based on the pedestrian trajectory; The second front-end device and the first front-end device jointly process the encrypted image to extract first image features, including: adjusting the speed of extracting the first image features based on the operating status of the first front-end device and the second front-end device.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the image feature processing method according to any one of claims 1 to 6 is implemented.

9. An electronic device, characterized in that: The electronic device comprises: a memory storing a computer program; a processor, communicatively connected to the memory, and configured to execute the image feature processing method according to any one of claims 1 to 6 when calling the computer program; A display is communicatively connected to the processor and the memory, and is used to display a GUI interaction interface related to the image feature processing method.

Images