Image recognition method, device, equipment and storage medium

Abstract

The present invention discloses an image recognition method, device, equipment and storage medium, the method comprising: receiving a panoramic image transmitted by a panoramic camera, and determining that the currently received panoramic image is an image to be recognized; performing computer visual recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; displaying the visual recognition result for supervision of the designated area; wherein the panoramic image is acquired by the panoramic camera for the designated area, and the visual recognition model is pre-trained using panoramic images acquired historically for the designated area. It can be seen that the present application realizes 360-degree all-round image acquisition of the area to be supervised based on the panoramic camera, which not only does not require the use of numerous cameras, but also can eliminate blind spots, thereby enabling effective supervision of the corresponding area.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and more specifically, to an image recognition method, device, equipment and storage medium. Background Art

[0002] At present, most of the known video surveillance systems are composed of a single camera with limited field of view. Due to the limitation of field of view, a large number of cameras are needed to cover an area, and there may still be blind spots, making it impossible to achieve effective supervision of the corresponding area. Summary of the invention

[0003] The purpose of the present invention is to provide an image recognition method, device, equipment and storage medium, which not only do not need to use a large number of cameras, but also can eliminate blind spots and achieve effective supervision of the corresponding area.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] An image recognition method, comprising:

[0006] Receiving a panoramic image transmitted by a panoramic camera, and determining that the currently received panoramic image is an image to be identified; wherein the panoramic image is acquired by the panoramic camera for a designated area;

[0007] Performing computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for the designated area;

[0008] The visual recognition results are displayed for monitoring the designated area.

[0009] Preferably, receiving a panoramic image transmitted by a panoramic camera includes:

[0010] Receive the panoramic image transmitted by the panoramic camera through the 5G network.

[0011] Preferably, using a visual recognition model to perform computer visual recognition on the image to be recognized to obtain a corresponding visual recognition result includes:

[0012] Performing equidistant segmentation on the image to be identified to obtain a corresponding plurality of sub-images to be identified;

[0013] Using the visual recognition model, simultaneously perform computer visual recognition on the plurality of sub-images to be recognized, respectively, to obtain a plurality of corresponding sub-visual recognition results;

[0014] The plurality of sub-visual recognition results are spliced ​​and filled according to the positions of the corresponding sub-images to be recognized in the image to be recognized, so as to obtain the visual recognition result.

[0015] Preferably, the image to be identified is segmented equidistantly to obtain a corresponding plurality of sub-images to be identified, including:

[0016] The image to be identified is equidistantly divided to obtain a plurality of corresponding sub-images to be identified, wherein the sub-images to be identified that are located at adjacent positions in the image to be identified have corresponding overlapping areas;

[0017] Correspondingly, after obtaining the corresponding multiple sub-visual recognition results, it also includes:

[0018] For multiple results corresponding to the same overlapping area in the sub-recognition results, only the results that meet the preset requirements are retained.

[0019] Preferably, after determining that the currently received panoramic image is the image to be identified, the method further includes:

[0020] Determine whether the image to be identified is an image that can be displayed normally. If so, execute the step of performing computer visual recognition on the image to be identified using a visual recognition model. If not, delete the image to be identified and continue to execute the step of receiving the panoramic image transmitted by the panoramic camera.

[0021] Preferably, the visual recognition model is trained in advance using panoramic images collected historically for the designated area, including:

[0022] Acquire multiple panoramic images captured by the panoramic camera for the designated area in the past as sample images;

[0023] The sample images are used to train a neural network model based on deep learning to obtain a corresponding visual recognition model.

[0024] Preferably, before using the sample image to train the neural network model based on deep learning, the method further includes:

[0025] Perform image enhancement processing on the sample image.

[0026] An image recognition device, comprising:

[0027] A receiving module, used to: receive a panoramic image transmitted by a panoramic camera, and determine that the currently received panoramic image is an image to be identified; wherein the panoramic image is acquired by the panoramic camera for a specified area;

[0028] A recognition module, used to: perform computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for the designated area;

[0029] The display module is used to display the visual recognition result for monitoring the designated area.

[0030] An image recognition device, comprising:

[0031] Memory for storing computer programs;

[0032] A processor is used to implement the steps of the image recognition method as described in any one of the above items when executing the computer program.

[0033] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above image recognition methods are implemented.

[0034] The present invention provides an image recognition method, device, equipment and storage medium, the method comprising: receiving a panoramic image transmitted by a panoramic camera, and determining that the currently received panoramic image is an image to be recognized; using a visual recognition model to perform computer visual recognition on the image to be recognized to obtain a corresponding visual recognition result; displaying the visual recognition result for supervision of the designated area; wherein the panoramic image is acquired by the panoramic camera for the designated area, and the visual recognition model is pre-trained using panoramic images acquired for the designated area in history. The present application utilizes a panoramic camera to realize the acquisition of panoramic images of the area to be supervised, and then realizes corresponding computer visual recognition based on the acquired panoramic image, and displays the visual recognition results of the computer visual recognition, so that the supervisor can realize corresponding supervision based on the displayed visual recognition results. It can be seen that the present application realizes 360-degree all-round image acquisition of the area to be supervised based on the panoramic camera, which not only does not require the use of many cameras, but also can eliminate blind spots, thereby realizing effective supervision of the corresponding area. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0036] Figure 1A flowchart of an image recognition method provided by an embodiment of the present invention;

[0037] Figure 2 A structural diagram of CSPDarknet53 in an image recognition method provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of data enhancement in an image recognition method provided by an embodiment of the present invention;

[0039] Figure 4 A flowchart for implementing an image recognition method according to an embodiment of the present invention;

[0040] Figure 5 A schematic diagram of the structure of an image recognition device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] See also Figure 1 , which shows a flow chart of an image recognition method provided by an embodiment of the present invention, which may include:

[0043] S11: receiving a panoramic image transmitted by a panoramic camera, and determining that the currently received panoramic image is an image to be identified; wherein the panoramic image is acquired by the panoramic camera for a designated area.

[0044] The execution subject of an image recognition method provided in an embodiment of the present invention may be a corresponding image recognition device or equipment, and the image recognition device or equipment may be collectively referred to as a supervision system. The embodiment of the present application specifically describes the execution subject of the image recognition method as a supervision system. Accordingly, in the embodiment of the present application, a module equipped with a camera and used to realize image acquisition of the supervised area using the camera may be referred to as a monitoring terminal.

[0045] It should be noted that the designated area is the area that needs to be supervised, and the camera installed on the monitoring terminal can be a panoramic camera, and the image collected by the panoramic camera can be called a panoramic image; the monitoring terminal uses the panoramic camera to collect images of the designated area, and then transmits the collected panoramic image to the supervision system, so that the supervision system can realize the corresponding image recognition for effective supervision and other operations. Among them, when the panoramic camera is used to realize the image collection of the designated area, it can ensure that the effective field of view of the camera in the horizontal and vertical directions is 360 degrees in all directions, thereby ensuring that the panoramic camera can realize the 360-degree full-scale image collection and corresponding supervision operations of the designated area, avoiding the shortcomings of the prior art that a large number of cameras are required and there may be blind spots. In addition, the panoramic camera in the embodiment of the present application can be specifically a VR device, which can be realized by using panoramic VR technology when realizing image collection. Panoramic VR is also called 3D real scene, which is an emerging rich media technology. It can use multiple single cameras to take real-scene photos, and after special splicing and processing, the 360-degree real scene is displayed in one picture, and can be operated and interacted with VR devices, so that the supervisor can remotely supervise the designated area from any perspective, thereby achieving the effect of removing blind spots in the supervised area in this way.

[0046] S12: Performing computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for a designated area.

[0047] After receiving the panoramic image captured and transmitted by the panoramic camera, the embodiment of the present application can use the pre-trained visual recognition model to perform computer vision recognition on the panoramic image. In the embodiment of the present application, the computer vision recognition of the panoramic image can specifically be target recognition on the panoramic image, so as to obtain the target to be recognized contained in the panoramic image. For example, if the panoramic image is an image of an intersection, the target can be a vehicle; if the panoramic image is an entrance or exit of a shopping mall, the target can be a person. Of course, other settings based on actual needs are within the scope of protection of the present invention. Target recognition has the same meaning as the corresponding concept in the prior art. Specifically, target recognition is the process of distinguishing a special target (or a type of target) from other targets (or other types of targets). It includes both the recognition of two very similar targets and the recognition of one type of target from other types of targets.

[0048] In addition, panoramic images taken by a panoramic camera of a designated area within a period of time before and closest to the current moment can be obtained in advance, and then these panoramic images can be used as sample images for training to obtain a visual recognition model for realizing computer visual recognition.

[0049] S13: Display the visual recognition results for monitoring the designated area.

[0050] The supervisor can perform operations such as supervision of the designated area through the displayed visual recognition results. It should also be noted that in a specific scenario, the designated area in this application can be a certain intersection, and the target to be identified can be the vehicle at the intersection; in this specific scenario, the embodiment of the present application can collect a panoramic image of the designated intersection through a panoramic camera, and then perform target recognition on the panoramic image to obtain the designated vehicle contained therein, that is, the visual recognition result, so that the supervisor can supervise the traffic conditions at the designated intersection through the visual recognition result; of course, it can also be applied to other specific application scenarios according to actual needs, all within the protection scope of the present invention.

[0051] This application uses a panoramic camera to collect panoramic images of the area to be supervised, and then implements corresponding computer vision recognition based on the collected panoramic images, and displays the visual recognition results of the computer vision recognition, so that the regulator can implement corresponding supervision based on the displayed visual recognition results. It can be seen that this application realizes 360-degree all-round image collection of the area to be supervised based on the panoramic camera, which not only does not require the use of many cameras, but also can eliminate blind spots, thereby achieving effective supervision of the corresponding area.

[0052] In addition, after the visual recognition model is pre-trained in the present application, the visual recognition model can be used offline to realize computer visual recognition of panoramic images at any time, thereby facilitating the rapid and easy realization of computer visual recognition of panoramic images.

[0053] An image recognition method provided by an embodiment of the present invention receives a panoramic image transmitted by a panoramic camera, and may include:

[0054] Receive panoramic images transmitted by the panoramic camera via the 5G network.

[0055] In the embodiment of the present application, the panoramic camera can collect video streams, and each frame of the collected video stream is a corresponding panoramic image. In a specific way, the communication between the monitoring terminal and the supervision system can be realized based on 5G communication technology. 5G (5th generation mobile networks) is the latest generation of cellular mobile communication technology, which has the characteristics of high data rate, reduced delay, energy saving, reduced cost, increased system capacity and support for large-scale device connection. Correspondingly, after the monitoring terminal collects the panoramic image of the specified area, it can transmit the panoramic image to the supervision system through the 5G network, that is, the present application uses the latest generation of 5G cellular mobile communication channels to solve the transmission problem of 4K / 8K video streams with large bandwidth and low latency, and realize high-speed and high-fidelity transmission of video streams; at the same time, due to the support of 5G communication technology, it can cope with ultra-high-resolution video stream transmission, which makes it possible to obtain better resolution for small targets during machine vision recognition as input to the supervision system.

[0056] An image recognition method provided by an embodiment of the present invention uses a visual recognition model to perform computer visual recognition on an image to be recognized to obtain a corresponding visual recognition result, which may include:

[0057] Performing equidistant segmentation on the image to be identified to obtain a corresponding plurality of sub-images to be identified;

[0058] Using the visual recognition model, computer visual recognition is performed on multiple sub-images to be recognized simultaneously to obtain corresponding multiple sub-visual recognition results;

[0059] The multiple sub-visual recognition results are spliced ​​and filled according to the positions of the corresponding sub-images to be recognized in the image to be recognized to obtain a visual recognition result.

[0060] It should be noted that in order to improve the recognition efficiency of computer vision recognition, the embodiment of the present application can perform equidistant segmentation on the image to be recognized, and the multiple image blocks obtained after the equidistant segmentation of the image to be recognized are all used as sub-images to be recognized, and then these multiple sub-images to be recognized are simultaneously input into multiple visual recognition models in a one-to-one correspondence. After the results output by these multiple visual recognition models are respectively sub-visual recognition results, these sub-visual recognition results are spliced ​​and filled according to the positions of their corresponding sub-images to be recognized in the image to be recognized, so as to obtain the final visual recognition result. Among them, the above-mentioned multiple visual recognition models are all trained in the same way, and all the obtained sub-visual recognition results are spliced ​​and filled, that is, each sub-visual recognition result is filled to the position of its corresponding sub-image to be recognized in the image to be recognized, and then these filled sub-visual recognition results are spliced ​​together in sequence according to the positions of their corresponding sub-images to be recognized in the image to be recognized, so that the overall visual recognition result can be obtained. For example, if the sub-image to be recognized corresponding to any sub-visual recognition result is located in the upper left corner of the image to be recognized, then the sub-visual recognition result is located in the upper left corner of the overall visual recognition result; if the sub-image to be recognized corresponding to any sub-visual recognition result is located in the upper right corner of the image to be recognized, then the sub-visual recognition result is located in the upper right corner of the overall visual recognition result; and so on, each sub-visual recognition result is located at the corresponding position of its corresponding sub-image to be recognized in the image to be recognized, and then the overall visual recognition result can be obtained.

[0061] In addition, in the embodiment of the present application, a process lock can be used to lock the process (which can be called a merge process) used to realize the splicing and filling of sub-visual recognition results, so as to prohibit it from performing normal operations of splicing and filling sub-visual recognition results, until all sub-visual recognition results are obtained and then the process lock is released, so that the above-mentioned merge process can splice and fill multiple sub-visual recognition results according to the position of the corresponding sub-image to be recognized in the image to be recognized, so as to avoid the incomplete visual recognition result obtained. In addition, each process that controls the sub-visual recognition results based on the sub-image to be recognized can be called a recognition process, and the recognition process can have a corresponding process shared variable. If any recognition process can obtain the corresponding sub-visual recognition result, the corresponding sub-visual recognition result is returned using the process shared variable, otherwise, it is returned empty to indicate that there is a problem with the sub-visual recognition result, indicating that the merge process does not need to wait for the arbitrary recognition process to return the sub-visual recognition result. It can be seen that in order to allow multiple sub-images to be recognized to be forward-conducted in the model without interfering with each other, the present application uses multiple processes to process different blocks of a large image in parallel, and uses process shared variables and process locks to ensure the stability of the program.

[0062] An image recognition method provided by an embodiment of the present invention performs equidistant segmentation on an image to be recognized to obtain a corresponding plurality of sub-images to be recognized, which may include:

[0063] The image to be identified is equidistantly divided to obtain a plurality of corresponding sub-images to be identified, and the sub-images to be identified located at adjacent positions in the image to be identified have corresponding overlapping areas;

[0064] Correspondingly, after obtaining the corresponding multiple sub-visual recognition results, the following may also be included:

[0065] For multiple results corresponding to the same overlapping area in the sub-recognition results, only the results that meet the preset requirements are retained.

[0066] It should be noted that the embodiment of the present application is aimed at ultra-high-resolution images to be identified, and while performing equidistant segmentation on the images to be identified, it can also ensure that there is a certain overlap between the areas obtained after segmentation; accordingly, computational visual recognition is performed on the segmented sub-images to be identified, and after detection and identification of all sub-images to be identified, splicing and filling are performed according to the positions in the images to be identified, and only the best result of the overlapping objects is retained for the overlapping areas therein, thereby further ensuring the accuracy of image recognition. In addition, after the visual recognition model performs computer visual recognition on any sub-image to be identified, the probability of whether there is a target to be detected at any position in the sub-image to be identified is obtained, and the best result retained is the result with the highest probability (i.e., the result that meets the preset requirements). Of course, other settings made according to actual needs are also within the scope of protection of the present invention.

[0067] An image recognition method provided by an embodiment of the present invention may further include, after determining that a currently received panoramic image is an image to be recognized:

[0068] Determine whether the image to be identified is an image that can be displayed normally. If so, execute the step of performing computer visual recognition on the image to be identified using the visual recognition model. If not, delete the image to be identified and continue to execute the step of receiving the panoramic image transmitted by the panoramic camera.

[0069] It should be noted that in the embodiment of the present application, before performing computer vision recognition on the image to be recognized, it can also be determined whether the image to be recognized is an image that can be displayed normally (such as a flower screen is an image that cannot be displayed normally). If so, valid information can be recognized from the image to be recognized, and the subsequent steps of performing computer vision recognition on it are performed. Otherwise, it means that valid information cannot be recognized from the image to be recognized, and the panoramic image can be directly deleted, thereby avoiding operations such as computer vision recognition of invalid images to be recognized, reducing waste of resources and time.

[0070] An image recognition method provided by an embodiment of the present invention, which uses panoramic images collected historically for a designated area to train a visual recognition model, may include:

[0071] Acquire multiple panoramic images captured by panoramic cameras for a specified area in history as sample images;

[0072] The sample images are used to train the neural network model based on deep learning to obtain the corresponding visual recognition model.

[0073] It should be noted that the embodiment of the present application can obtain multiple panoramic images taken by the panoramic camera of the specified area before the current moment and within a period of time closest to the current moment, and then use these panoramic images to train the neural network model based on deep learning to obtain the corresponding visual recognition model. It can be seen that the present application implements image recognition based on the neural network of deep learning, which can ensure the accuracy of image recognition.

[0074] Specifically, for the deep learning-based neural network model used in the embodiments of the present application, the head part of the neural network model is retained in the algorithm used in the network structure, and the backbone network is modified to CSPDarknet53 (such as Figure 2 As shown), the idea of ​​SPP is used to expand the receptive field, and PANet is used as the neck part. Among them, CSPDarknet53 contains 29 convolutional layers, a receptive field of 725*725 and 27.6M parameters; the original PANet uses Addition when fusion, and the addition operation is changed to multiplication in the algorithm of the embodiment of the present application; Cross-mini-batch Normal, CBN normalizes the results of the current and the previous 3 mini-batches, and the neural network model used is improved on the basis of CBN; modified SAM, from SAM's space-by-space attention to point-by-point attention. In order to identify targets more quickly and efficiently, the provided lightweight model can be used.

[0075] In the process of target detection and recognition, the regression box, confidence and classification loss value of the algorithm are used as the targets that need to be converged. In the post-processing of the algorithm, some unreliable results are filtered out by adjusting the IOU and confidence thresholds. For the loss of the regression box, in order to detect objects of different sizes under different fields of view, three groups of reference boxes under the field of view are designed, which are responsible for 8x downsampling, 16x downsampling and 32x downsampling respectively. The center of the result box output by the network is compared with the corresponding reference box. CIoU loss is used to make more detailed measurements based on the distance between the center points of the two boxes, such as overlapping area and aspect ratio. There is a formula:

[0076]

[0077]

[0078] Among them, v measures the distance of the aspect ratio of the two boxes, α is a balance coefficient, the value of which is determined by IoU and v, and w p and h p They are all normalized values; type loss, in target detection, generally uses cross entropy loss (binary_crossentropy), calculates the cross entropy loss for each category and then sums them; confidence loss also uses binary cross entropy loss.

[0079] An image recognition method provided by an embodiment of the present invention may further include, before training a neural network model based on deep learning using sample images:

[0080] Perform image enhancement processing on the sample image.

[0081] In order to further enhance the recognition accuracy of the visual recognition model, this application can use Mosaic data enhancement (such as Figure 3 As shown in the figure, four sample images are stitched together into one image for training, which is equivalent to increasing the mini-batch in disguise. The following methods can also be used to enhance the data of sample images, such as image perturbation, brightness change, contrast change, saturation change, hue change, noise addition, random scaling, random cropping, flipping, rotation, and random erasing, to enhance the robustness and generalization performance of the visual recognition model in the production environment.

[0082] In a specific implementation, an image recognition method provided by an embodiment of the present invention can be as follows: Figure 4As shown, whether the segmentation is wrong is to judge whether the image can be displayed normally. It can be seen that the present invention adopts VR panoramic technology to remove blind spots. The use of panoramic cameras can ensure that the effective field of view of the camera in the horizontal and vertical directions is 360 degrees in all directions. With the support of 5G communication technology, high-speed and high-fidelity transmission of video streams can be achieved. At the same time, due to the support of 5G communication technology, it is possible to cope with ultra-high-resolution video stream transmission, which makes it possible to obtain better resolution for small targets through processing as input to the supervision system; through the interaction of VR devices, supervisors can remotely supervise the supervision site from any perspective.

[0083] The embodiment of the present invention also provides an image recognition device, such as Figure 5 As shown, it may include:

[0084] The receiving module 11 is used to: receive a panoramic image transmitted by a panoramic camera, and determine that the currently received panoramic image is an image to be identified; the panoramic image is acquired by the panoramic camera for a specified area;

[0085] The recognition module 12 is used to: perform computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for a designated area;

[0086] The display module 13 is used to display the visual recognition result for monitoring the designated area.

[0087] An image recognition device provided by an embodiment of the present invention, wherein the receiving module may include:

[0088] The receiving unit is used to receive the panoramic image transmitted by the panoramic camera through the 5G network.

[0089] An embodiment of the present invention provides an image recognition device, wherein the recognition module may include:

[0090] The recognition unit is used to: equidistantly divide the image to be recognized to obtain a corresponding plurality of sub-images to be recognized; use a visual recognition model to simultaneously perform computer visual recognition on the multiple sub-images to be recognized to obtain a corresponding plurality of sub-visual recognition results; splice and fill the multiple sub-visual recognition results according to the positions of the corresponding sub-images to be recognized in the image to be recognized to obtain a visual recognition result.

[0091] An image recognition device provided by an embodiment of the present invention, wherein the recognition unit may include: a recognition sub-unit, which is used to: equidistantly divide the image to be recognized to obtain a corresponding plurality of sub-images to be recognized, and the sub-images to be recognized located at adjacent positions in the image to be recognized have corresponding overlapping areas;

[0092] Correspondingly, the image recognition device may further include: a processing module, which is used to: after obtaining the corresponding multiple sub-visual recognition results, for multiple results corresponding to the same overlapping area in the sub-recognition results, only retain the results that meet the preset requirements among the multiple results.

[0093] An image recognition device provided by an embodiment of the present invention may further include:

[0094] The judgment module is used to: after determining that the currently received panoramic image is the image to be recognized, judge whether the image to be recognized is an image that can be displayed normally. If so, execute the step of performing computer visual recognition on the image to be recognized using the visual recognition model; if not, delete the image to be recognized and continue to execute the step of receiving the panoramic image transmitted by the panoramic camera.

[0095] An image recognition device provided by an embodiment of the present invention may further include:

[0096] The training module is used to: obtain multiple panoramic images historically acquired by panoramic cameras for a specified area as sample images; use the sample images to train a neural network model based on deep learning to obtain a corresponding visual recognition model.

[0097] An image recognition device provided by an embodiment of the present invention may further include:

[0098] The preprocessing module is used to perform image enhancement processing on the sample images before using the sample images to train the neural network model based on deep learning.

[0099] The embodiment of the present invention further provides an image recognition device, which may include:

[0100] Memory for storing computer programs;

[0101] A processor is used to implement the steps of any of the above image recognition methods when executing a computer program.

[0102] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above image recognition methods can be implemented.

[0103] It should be noted that the description of the relevant parts of an image recognition device, equipment and storage medium provided by the embodiment of the present invention can refer to the detailed description of the corresponding parts in an image recognition method provided by the embodiment of the present invention, which will not be repeated here. In addition, the parts of the above technical solutions provided by the embodiment of the present invention that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0104] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An image recognition method, characterized in that: include: Receiving a panoramic image transmitted by a panoramic camera, and determining that the currently received panoramic image is an image to be identified; wherein the panoramic image is acquired by the panoramic camera for a designated area; and the panoramic camera is a VR device; Performing computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for the designated area; Displaying the visual recognition result for monitoring the designated area; The receiving of the panoramic image transmitted by the panoramic camera includes: Receiving a panoramic image transmitted by the panoramic camera via a 5G network; Perform computer vision recognition on the image to be recognized using a visual recognition model to obtain corresponding visual recognition results, including: Performing equidistant segmentation on the image to be identified to obtain a corresponding plurality of sub-images to be identified; Using the visual recognition model and the process lock, computer visual recognition is performed on the plurality of sub-images to be recognized at the same time, until a plurality of corresponding sub-visual recognition results are determined based on the obtained plurality of process shared variables, and the process lock is released; the process lock is used to lock the process of splicing and filling the sub-visual recognition results; The plurality of sub-visual recognition results are spliced ​​and filled according to the positions of the corresponding sub-images to be recognized in the image to be recognized, so as to obtain the visual recognition result.

2. The method according to claim 1, characterized in that The image to be identified is segmented equidistantly to obtain a plurality of corresponding sub-images to be identified, including: The image to be identified is equidistantly divided to obtain a plurality of corresponding sub-images to be identified, wherein the sub-images to be identified that are located at adjacent positions in the image to be identified have corresponding overlapping areas; Correspondingly, after obtaining the corresponding multiple sub-visual recognition results, it also includes: For multiple results corresponding to the same overlapping area in the sub-recognition results, only the results that meet the preset requirements are retained.

3. The method according to claim 2, characterized in that After determining that the currently received panoramic image is an image to be identified, the method further includes: Determine whether the image to be identified is an image that can be displayed normally. If so, execute the step of performing computer visual recognition on the image to be identified using a visual recognition model. If not, delete the image to be identified and continue to execute the step of receiving the panoramic image transmitted by the panoramic camera.

4. The method according to claim 3, characterized in that The visual recognition model is trained in advance using panoramic images collected historically for the designated area, including: Acquire multiple panoramic images captured by the panoramic camera for the designated area in the past as sample images; The sample images are used to train a neural network model based on deep learning to obtain a corresponding visual recognition model.

5. The method according to claim 4, characterized in that Before using the sample image to train the neural network model based on deep learning, the method further includes: Perform image enhancement processing on the sample image.

6. An image recognition device, characterized in that: include: A receiving module, used to: receive a panoramic image transmitted by a panoramic camera, and determine that the currently received panoramic image is an image to be identified; wherein the panoramic image is acquired by the panoramic camera for a specified area; and the panoramic camera is a VR device; A recognition module, used to: perform computer vision recognition on the image to be recognized using a visual recognition model to obtain a corresponding visual recognition result; wherein the visual recognition model is pre-trained using panoramic images collected historically for the designated area; A display module, used to: display the visual recognition result for monitoring the designated area; Wherein, the receiving module includes: A receiving unit, used to: receive a panoramic image transmitted by a panoramic camera via a 5G network; Identification modules may include: The recognition unit is used to: equidistantly divide the image to be recognized to obtain a corresponding plurality of sub-images to be recognized; use the visual recognition model and the process lock to simultaneously perform computer visual recognition on the plurality of sub-images to be recognized, until a corresponding plurality of sub-visual recognition results are determined based on the obtained plurality of process shared variables, and the process lock is released; the process lock is used to lock the process for splicing and filling the sub-visual recognition results; and splice and fill the plurality of sub-visual recognition results according to the positions of the corresponding sub-images to be recognized in the image to be recognized to obtain the visual recognition result.

7. An image recognition device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the image recognition method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the image recognition method according to any one of claims 1 to 5 are implemented.

Images