Target detection method and apparatus

Abstract

The application provides a target detection method and device, and relates to the field of information technology.The method comprises the following steps: an edge computing platform extracts features of a target image, and determines a contour image of the target image; the edge computing platform determines a target candidate region according to the contour image; and the edge computing platform sends the target candidate region and the contour image to a cloud platform, and the target candidate region and the contour image are used for target detection of the target image.The target detection method and device provided in the application embodiment realize image acquisition, feature extraction, target candidate region identification through the edge computing platform, and the cloud platform performs target detection reasoning and training of a detection model, so that the advantages of rich cloud models, rich computing power and storage resources are fully utilized, efficient target detection is provided for the edge computing platform, and the image processing capability of the edge computing platform is utilized, so that the consumption of network bandwidth is reduced, and fast target detection is realized under the condition of a low-bandwidth network.

Description

Technical Field

[0001] This application relates to the field of information technology, specifically to a target detection method and apparatus. Background Technology

[0002] Current object detection techniques include: AlexNet, a deep convolutional neural network (DCNN) proposed by Krizhevsky et al., achieved higher image classification accuracy in visual recognition challenges, thus shifting the focus of computer vision application research to deep learning methods. This method has the following drawbacks:

[0003] Algorithms such as RCNN integrate feature extraction, region recognition, and inference, requiring significant computing power and thus cannot be deployed on IoT terminals with limited computing capabilities. If all acquisition devices transmit images or videos to the cloud for feature extraction and object recognition, it relies on high-bandwidth, low-latency 5G networks, which are not suitable for industrial scenarios. Summary of the Invention

[0004] This application provides a target detection method to solve the technical problem of achieving fast target detection under low bandwidth network conditions.

[0005] In a first aspect, embodiments of this application provide a target detection method, including:

[0006] The edge computing platform extracts features from the target image to determine the contour image of the target image;

[0007] The edge computing platform determines the target candidate region based on the contour image;

[0008] The edge computing platform sends the target candidate region and the contour image to the cloud platform, and the target candidate region and the contour image are used to perform target detection on the target image.

[0009] In one embodiment, the edge computing platform performs feature extraction on the target image to determine the contour image of the target image, including:

[0010] The edge computing platform extracts the contour features of the target image and determines the feature image of the target image;

[0011] The edge computing platform divides the feature image into multiple target units of target pixel value size, and determines the contour image based on the multiple target units.

[0012] In one embodiment, after the edge computing platform extracts the contour features of the target image, the method further includes:

[0013] The edge computing platform determines the difference regions between the target image and its adjacent frames in the target video;

[0014] If the area of ​​the difference region does not exceed an area threshold, the edge computing platform extracts the contour features of the difference region.

[0015] The edge computing platform superimposes the contour features of the difference region onto the contour features of the target image, and determines the feature image of the superimposed target image based on the superimposed contour features.

[0016] In one embodiment, the edge computing platform determines a target candidate region based on the contour image, including:

[0017] The edge computing platform determines the target color block based on the target unit in the contour image;

[0018] The difference in grayscale values ​​between any two adjacent target units within the target color block is less than the grayscale value threshold.

[0019] The edge computing platform determines the target candidate region based on the vertex coordinates of the target color block, as well as the length and width of the target color block.

[0020] In one embodiment, the edge computing platform sends the target candidate region and the contour image to the cloud platform, including:

[0021] The edge computing platform determines the weight corresponding to the level of the access service based on the preset service level.

[0022] The edge computing platform determines the bandwidth occupied by the access service level based on the weight.

[0023] The edge computing platform sends the target candidate region and the contour image to the cloud platform based on the bandwidth.

[0024] Secondly, embodiments of this application provide a target detection method, including:

[0025] The cloud platform receives target candidate region and contour images sent by the edge computing platform;

[0026] The cloud platform performs target detection on the target image based on the target candidate region and the contour image;

[0027] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0028] The target candidate region is determined by the edge computing platform based on the contour image.

[0029] Thirdly, embodiments of this application provide a target detection device, comprising:

[0030] The first determining module is used by the edge computing platform to extract features from the target image and determine the contour image of the target image;

[0031] The second determining module is used by the edge computing platform to determine the target candidate region based on the contour image;

[0032] The sending module is used by the edge computing platform to send the target candidate region and the contour image to the cloud platform, wherein the target candidate region and the contour image are used to perform target detection on the target image.

[0033] Fourthly, embodiments of this application provide a target detection device, comprising:

[0034] The receiving module is used by the cloud platform to receive target candidate region and contour images sent by the edge computing platform;

[0035] The detection module is used by the cloud platform to perform target detection on the target image based on the target candidate region and the contour image;

[0036] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0037] The target candidate region is determined by the edge computing platform based on the contour image.

[0038] Fifthly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the target detection method described in the first or second aspect.

[0039] In a sixth aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the target detection method described in the first or second aspect.

[0040] The target detection method, apparatus, electronic device, and computer program product provided in this application embodiment perform image acquisition, feature extraction, and target candidate region recognition through an edge computing platform, and target detection inference and detection model training through a cloud platform. It makes full use of the advantages of the cloud, such as rich models, abundant computing power, and storage resources, to provide high-efficiency target detection for the edge computing platform. At the same time, it utilizes the image processing capabilities of the edge computing platform to reduce the consumption of network bandwidth and achieve fast target detection under low bandwidth network conditions. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is one of the flowcharts of the target detection method provided in the embodiments of this application;

[0043] Figure 2 This is a second schematic flowchart of the target detection method provided in the embodiments of this application;

[0044] Figure 3 This is one of the structural schematic diagrams of the target detection device provided in the embodiments of this application;

[0045] Figure 4 This is a second schematic diagram of the target detection device provided in the embodiments of this application;

[0046] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] Figure 1 This is one of the flowcharts illustrating the target detection method provided in this application. (Refer to...) Figure 1 The target detection method provided in this application embodiment may include:

[0049] Step 110: The edge computing platform extracts features from the target image to determine the contour image of the target image;

[0050] Step 120: The edge computing platform determines the target candidate region based on the contour image;

[0051] Step 130: The edge computing platform sends the target candidate region and contour image to the cloud platform. The target candidate region and contour image are used to perform target detection on the target image.

[0052] It should be noted that the execution entity of the target detection method provided in this application embodiment can be an edge computing platform. Edge computing refers to an open platform that integrates network, computing, storage, and application core capabilities on the side closest to the object or data source, providing the nearest end service. Its applications are initiated on the edge side, resulting in faster network service response and meeting the basic needs of industries in real-time business, application intelligence, security, and privacy protection. Edge computing is located between physical entities and industrial connections, or on top of physical entities. Cloud computing can still access the historical data of edge computing.

[0053] Optionally, in step 110, the edge computing platform performs feature extraction on the target image to determine the contour image of the target image.

[0054] Edge computing platforms acquire images or videos through image acquisition devices and extract features from the acquired target images. The features of the target images can include contour features, color boundary features, and proportional features between different regions. The edge computing platform extracts features from the acquired target images to determine the contour images in the target images.

[0055] In step 120, the edge computing platform uses the contour image determined in the target image as the region where the contour image is located as the target candidate region, i.e., the region to be detected. There can be one or more target candidate regions.

[0056] In step 130, the edge computing platform sends the target candidate region and contour image to the cloud platform. The target candidate region and contour image are used to perform target detection on the target image.

[0057] The edge computing platform sends the target candidate regions and contour images from the target image to the cloud platform. After receiving the information from the edge computing platform, the cloud platform executes the deep neural network model training and inference tasks. Because the edge computing platform has already preprocessed the image, the amount of computation during training and inference can be significantly reduced, thereby enabling rapid target detection in the target image.

[0058] Furthermore, after acquiring the target image, the edge computing platform can identify common objects within the image, such as monitors, tires, and people. Based on the size of these objects, it infers the image scaling ratio, corresponding to the image capture distance. The image is then scaled according to this ratio to fit the model size, resulting in a single-scale image that effectively enhances the generalization ability of the neural network.

[0059] The target detection method provided in this application embodiment performs image acquisition, feature extraction, and target candidate region identification through an edge computing platform, and target detection inference and detection model training through a cloud platform. It fully utilizes the advantages of the cloud platform, such as rich models, abundant computing power and storage resources, to provide high-efficiency target detection for the edge computing platform. At the same time, it utilizes the image processing capabilities of the edge computing platform to reduce the consumption of network bandwidth and achieve fast target detection under low bandwidth network conditions.

[0060] In one embodiment, the edge computing platform performs feature extraction on the target image to determine the contour image of the target image, including:

[0061] The edge computing platform extracts the contour features of the target image to determine the feature image of the target image;

[0062] The edge computing platform divides the feature image into target units of multiple target pixel values ​​and determines the contour image based on these multiple target units.

[0063] Optionally, the edge computing platform first performs grayscale processing on the acquired images, for example:

[0064] Grayx,y=Rx,y*K1+Gx,y*K2+Bx,y*K3+K4

[0065] Where x and y are coordinates, R is the red component, G is the green component, and B is the blue component;

[0066] The conversion coefficients K1, K2, K3, and K4 can take the following values: K1 = 0.3, K2 = 0.59, K3 = 0.11, and K4 = 0.5.

[0067] By using the converted grayscale value as the z-coordinate, the grayscale image becomes a three-dimensional image. The edge computing platform can capture contour information based on the gradient data of each pixel, thereby determining the feature image of the target image.

[0068] Edge computing platforms divide feature images into target units of multiple target pixel values, such as 16*16 or 8*8 small units, which can enhance image feature information to determine the contour image, while also significantly compressing the image size.

[0069] The target detection method provided in this application uses an edge computing platform to extract features from an image and divides the feature image into multiple target units of target pixel value size. This enhances the image feature information while significantly compressing the image size and improving computational efficiency.

[0070] In one embodiment, after the edge computing platform extracts the contour features of the target image, the method further includes:

[0071] The edge computing platform identifies the regions of difference between the target image and adjacent frames in the target video;

[0072] If the area of ​​the difference region does not exceed the area threshold, the edge computing platform extracts the contour features of the difference region;

[0073] The edge computing platform overlays the contour features of the difference region onto the contour features of the target image, and determines the feature image of the overlaid target image based on the overlaid contour features.

[0074] Optionally, when the target image is acquired from video, the differences between adjacent frames are usually minimal. The edge computing platform identifies the difference regions between the target image and its adjacent frames. If the area of ​​the difference region does not exceed an area threshold (e.g., the area enclosed by the contours of the difference regions divided by the area of ​​the entire image is less than or equal to one-quarter), the edge computing platform extracts the contour features of the difference regions. These contour features are then superimposed onto the contour features of the target image, and based on the superimposed contour features, the feature image of the superimposed target image is determined.

[0075] Understandably, if the area of ​​the difference region exceeds the area threshold, for example, if the area enclosed by the outline of the difference region divided by the area of ​​the entire graphic is greater than one-quarter, then the outline information of the complete image is recalculated and the memory cache data is refreshed.

[0076] The target detection method provided in this application embodiment extracts the features of the difference regions of adjacent frame images through an edge computing platform and then superimposes them onto the cached graphic feature information, thereby reducing the amount of data input and further improving the computational efficiency.

[0077] In one embodiment, the edge computing platform determines a target candidate region based on a contour image, including:

[0078] The edge computing platform determines the target color block based on the target unit in the contour image;

[0079] The difference in grayscale values ​​between any two adjacent target units within the target color block is less than the grayscale value threshold;

[0080] The edge computing platform determines the target candidate region based on the vertex coordinates of the target color block, as well as the length and width of the target color block.

[0081] Alternatively, traditional algorithms for finding target regions include sliding window and candidate region algorithms. In real-world applications, targets may be occluded, contain each other, or vary in size due to distance, leading to low efficiency of traditional algorithms.

[0082] The target candidate region can be determined in the following ways according to the embodiments of this application:

[0083] Randomly select the smallest unit and search outwards to the eight adjacent units. If the deviation of the Gray x and y values ​​is less than 5%, it is considered to be the same color block. Then, using the units in the same color block as the center, continue the recursive search until the area of ​​the same color block no longer increases. All small units within the same color block are then identified as the target color block.

[0084] If there are still cells in the image that have not been colored, continue to randomly select a cell and repeat the above search and coloring steps until all cells are colored.

[0085] Starting with a cell with gradient feature information randomly selected by the edge computing platform, the cell size can gradually decrease and converge. Through this optimization, the search efficiency is significantly improved.

[0086] The edge computing platform determines candidate regions for targets based on the vertex coordinates of the target color patch, as well as the length and width of the target color patch. For example, the minimum x / y and maximum w / h of the target color patch can be used as a candidate region. The origin of the image is at the top left corner, the minimum x / y can be the coordinates of the top left corner of a rectangle that can cover the feature contour, and the maximum w / h can be the width and height of the rectangle that can cover the feature contour.

[0087] The target detection method provided in this application embodiment randomly selects a unit with gradient feature information through an edge computing platform, and determines the target candidate region based on the unit, which effectively improves the search efficiency.

[0088] In one embodiment, the edge computing platform sends a target candidate region and contour image to the cloud platform, including:

[0089] The edge computing platform determines the weight corresponding to the level of the access service based on the preset service level;

[0090] The edge computing platform determines the bandwidth usage level of the access service based on its weight.

[0091] The edge computing platform sends target candidate regions and contour images to the cloud platform based on bandwidth.

[0092] Optionally, the edge computing platform packages the processed image, target region parameters R1(x1,y1,w1,h1) and R2(x2,y2,w2,h2), local detail images, and other data together, compresses them, and sends them to the cloud. Here, x and y are the top-left coordinates of the target region; w and h are the width and height of the target region. If the image contains text information, the recognition results are also packaged and sent to the cloud.

[0093] Edge computing platforms classify connected services according to preset service levels, such as autonomous driving, security, and intelligent transportation.

[0094] For example, if the service levels are divided into SLevel1 to SLeveln from high to low, and the corresponding weights are SW1 to SWn, then the bandwidth occupied by each service level is SWi / ∑SW.

[0095] The transmission order is from high to low. If there are remaining bandwidth resources for high-priority services, they will be allocated to low-priority services. This ensures that high-priority services can occupy bandwidth first without blocking the response of low-priority services.

[0096] The target detection method provided in this application provides a hierarchical classification of access services through an edge computing platform. By determining the bandwidth occupied by each level of service according to its weight, it can satisfy the requirement that higher-level services occupy bandwidth first without blocking the response of lower-level services.

[0097] Figure 2 This is a second schematic flowchart illustrating the target detection method provided in this application's embodiments. (Refer to...) Figure 2 The target detection method provided in this application embodiment may include:

[0098] Step 210: The cloud platform receives the target candidate region and contour image sent by the edge computing platform;

[0099] Step 220: The cloud platform performs target detection on the target image based on the target candidate region and contour image;

[0100] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0101] The target candidate region is determined by the edge computing platform based on the contour image.

[0102] It should be noted that the execution entity of the target detection method provided in this application embodiment can be a cloud platform. Edge computing refers to an open platform that integrates network, computing, storage, and application core capabilities on the side closest to the object or data source, providing the nearest end service. Its applications are initiated on the edge side, resulting in faster network service response and meeting the basic needs of industries in real-time business, application intelligence, security, and privacy protection. Edge computing is located between physical entities and industrial connections, or on top of physical entities. Cloud computing can still access the historical data of edge computing.

[0103] Optionally, the edge computing platform acquires images or videos through image acquisition devices and extracts features from the acquired target images. The features of the target images may include contour features, color boundary features, and proportional features between different regions. The edge computing platform extracts features from the acquired target images to determine the contour images in the target images.

[0104] The edge computing platform uses the contour image determined in the target image as the region where the contour image is located as the target candidate region, i.e. the region to be detected. There can be one or more target candidate regions.

[0105] The edge computing platform sends the target candidate regions and contour images from the target image to the cloud platform. After receiving the information from the edge computing platform, the cloud platform executes the deep neural network model training and inference tasks. Because the edge computing platform has already preprocessed the image, the amount of computation during training and inference can be significantly reduced, thereby enabling rapid target detection in the target image.

[0106] Furthermore, after acquiring the target image, the edge computing platform can identify common objects within the image, such as monitors, tires, and people. Based on the size of these objects, it infers the image scaling ratio, corresponding to the image capture distance. The image is then scaled according to this ratio to fit the model size, resulting in a single-scale image that effectively enhances the generalization ability of the neural network.

[0107] The target detection method provided in this application embodiment performs image acquisition, feature extraction, and target candidate region identification through an edge computing platform, and target detection inference and detection model training through a cloud platform. It fully utilizes the advantages of the cloud platform, such as rich models, abundant computing power and storage resources, to provide high-efficiency target detection for the edge computing platform. At the same time, it utilizes the image processing capabilities of the edge computing platform to reduce the consumption of network bandwidth and achieve fast target detection under low bandwidth network conditions.

[0108] In one embodiment, the target detection method provided in this application may include:

[0109] Step 1: The edge computing platform acquires images or videos through image acquisition devices;

[0110] Step 2: The edge computing platform filters non-feature information in the image and detects target candidate regions. The specific process is as follows:

[0111] Step 2.1: The YD-Cat-Algorithm algorithm designed in this application first performs grayscale processing on the acquired image;

[0112] Step 2.2: Take the converted grayscale as the z-coordinate, then the grayscale image is a three-dimensional image. The contour information can be captured from the gradient data of each pixel. After dividing the image into 16*16 small units, the image feature information is enhanced while the image size is greatly compressed.

[0113] This application embodiment caches graphic feature information in memory, then extracts the features of the difference regions of adjacent images, and then superimposes them onto the cached graphic feature information in memory, which greatly improves the computational efficiency; if the difference regions of the images exceed the threshold, the contour information of the complete image is recalculated and the cached data in memory is refreshed.

[0114] This application embodiment finds some common objects in the image, infers the scaling ratio of the image based on the size of these objects, and calculates the image scaling ratio corresponding to the shooting distance. The image is then scaled according to this ratio to fit the model size. The resulting image is of a single scale, thereby effectively improving the generalization ability of the neural network.

[0115] Step 2.3: This embodiment of the application adopts the YD-Cat-Algorithm algorithm, that is, randomly select a smallest unit, and search for 8 adjacent units from it. If the deviation of the Gray x and y values ​​is less than 5%, it is considered to be the same color block. Then, recursively search for units with the same color block as the center until the area of ​​the same color block no longer increases. If there are still units in the image that have not been colored, randomly select a unit and repeat the above search and coloring steps. The minimum x / y and maximum w / h of the same color block is a target area (to be detected).

[0116] Step 3: The edge computing platform packages and compresses the processed image, target region parameters R1(x1,y1,w1,h1), R2(x2,y2,w2,h2), local detail images, etc., and sends them to the cloud. Here, x / y is the upper left coordinate point of the target region, and w / h is the width and height of the target region.

[0117] Step 3.1: If the image contains text information, package the recognition results together and send them to the cloud;

[0118] Step 3.2: The edge computing platform classifies the access services. Assuming that the service levels are divided into SLevel1 to SLeveln from high to low, and the corresponding weights are SW1 to SWn, the bandwidth occupied by each level of service is SWi / ∑SW. The sending order is from high to low. If there are remaining bandwidth resources for high-level services, they will be allocated to low-level services. This can satisfy the requirement that high-level services occupy bandwidth first without blocking the response of low-level services.

[0119] Step 4: After receiving the above information, the cloud platform executes the deep neural network model training and inference tasks. Because the edge computing platform has already preprocessed the images, the computational load for training and inference is significantly reduced. The cloud-based object detection training and inference model uses the YOLO algorithm, and the sample matching algorithm is optimized as follows:

[0120] Optimization 1: First, calculate the derivative of each cell, the derivative of the cells in the target area that are located on the contour, and then calculate the variance of this set of derivatives. This can reduce the number of sample matches and avoid overfitting caused by symmetrical contours.

[0121] Optimization 2: ROI filtering. If the aspect ratio of the target region is greater than a threshold (e.g., 10), it is considered an invalid region and is discarded directly to improve computational efficiency.

[0122] Step 5: The cloud returns the target detection results to the edge computing platform, facilitating the edge computing platform to continue subsequent business operations;

[0123] Step 6: If the target detection result in the cloud is lower than the set threshold, the original image data is retrieved from the edge computing platform again for inference, or it is manually labeled and used as input for model training.

[0124] According to the target detection method provided in this application, the edge computing platform is responsible for image acquisition, image feature extraction, target candidate region annotation, and image compression, while the cloud computing platform is responsible for model training, target detection, and inference model distribution (to all edge computing platforms), thereby achieving dynamic expansion of the detection capabilities of the edge computing platform. That is, a scheme of dynamically loading the inference model onto the edge computing platform is adopted to achieve dynamic expansion of detection capabilities. The YD-Cat-Algorithm algorithm is used, achieving faster target detection efficiency.

[0125] In IoT applications with low computing power and low bandwidth, distributed target detection systems have higher target detection accuracy. (1) Among 10,000 randomly sampled images, a total of 75,926 targets were detected, with a target detection accuracy of 99.72%; (2) Under the bandwidth of 4G network (1.5MB / s to 3MB / s), the average detection time is 735ms, which is more than 70% shorter than the average detection time of 2,650ms in the cloud.

[0126] The target detection device provided in the embodiments of this application is described below. The target detection device described below can be referred to in correspondence with the target detection method described above.

[0127] Figure 3 This is one of the target detection devices provided in the embodiments of this application, referred to... Figure 3 The target detection device provided in this application embodiment may include:

[0128] The first determining module 310 is used by the edge computing platform to extract features from the target image and determine the contour image of the target image;

[0129] The second determining module 320 is used by the edge computing platform to determine the target candidate region based on the contour image;

[0130] The sending module 330 is used by the edge computing platform to send the target candidate region and contour image to the cloud platform. The target candidate region and contour image are used to perform target detection on the target image.

[0131] The target detection device provided in this application embodiment performs image acquisition, feature extraction, and target candidate region recognition through an edge computing platform, and target detection inference and detection model training through a cloud platform. It fully utilizes the advantages of the cloud platform, such as rich models, abundant computing power and storage resources, to provide high-efficiency target detection for the edge computing platform. At the same time, it utilizes the image processing capabilities of the edge computing platform to reduce the consumption of network bandwidth and achieve fast target detection under low bandwidth network conditions.

[0132] In one embodiment, the first determining module 310 is specifically used for:

[0133] The edge computing platform extracts the contour features of the target image to determine the feature image of the target image;

[0134] The edge computing platform divides the feature image into target units of multiple target pixel values ​​and determines the contour image based on these multiple target units.

[0135] In one embodiment, after the edge computing platform extracts the contour features of the target image, the method further includes:

[0136] The edge computing platform identifies the regions of difference between the target image and adjacent frames in the target video;

[0137] If the area of ​​the difference region does not exceed the area threshold, the edge computing platform extracts the contour features of the difference region;

[0138] The edge computing platform overlays the contour features of the difference region onto the contour features of the target image, and determines the feature image of the overlaid target image based on the overlaid contour features.

[0139] In one embodiment, the second determining module 320 is specifically used for:

[0140] The edge computing platform determines the target color block based on the target unit in the contour image;

[0141] The difference in grayscale values ​​between any two adjacent target units within the target color block is less than the grayscale value threshold;

[0142] The edge computing platform determines the target candidate region based on the vertex coordinates of the target color block, as well as the length and width of the target color block.

[0143] In one embodiment, the sending module 330 is specifically used for:

[0144] The edge computing platform determines the weight corresponding to the level of the access service based on the preset service level;

[0145] The edge computing platform determines the bandwidth usage level of the access service based on its weight.

[0146] The edge computing platform sends target candidate regions and contour images to the cloud platform based on bandwidth.

[0147] Figure 4 This is the second target detection device provided in the embodiments of this application, referred to... Figure 4 The target detection device provided in this application embodiment may include:

[0148] The receiving module 410 is used for the cloud platform to receive the target candidate region and contour image sent by the edge computing platform;

[0149] The detection module 420 is used by the cloud platform to perform target detection on the target image based on the target candidate region and contour image;

[0150] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0151] The target candidate region is determined by the edge computing platform based on the contour image.

[0152] The target detection device provided in this application embodiment performs image acquisition, feature extraction, and target candidate region recognition through an edge computing platform, and target detection inference and detection model training through a cloud platform. It fully utilizes the advantages of the cloud platform, such as rich models, abundant computing power and storage resources, to provide high-efficiency target detection for the edge computing platform. At the same time, it utilizes the image processing capabilities of the edge computing platform to reduce the consumption of network bandwidth and achieve fast target detection under low bandwidth network conditions.

[0153] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call a computer program stored in the memory 530 to execute a target detection method, such as including:

[0154] The edge computing platform extracts features from the target image to determine its contour image;

[0155] The edge computing platform determines the target candidate region based on the contour image;

[0156] The edge computing platform sends target candidate regions and contour images to the cloud platform, which are used to perform target detection on the target image.

[0157] or

[0158] The cloud platform receives target candidate region and contour images sent by the edge computing platform;

[0159] The cloud platform performs target detection on the target image based on the target candidate region and contour image;

[0160] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0161] The target candidate region is determined by the edge computing platform based on the contour image.

[0162] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the target detection methods provided in the above embodiments, such as including:

[0164] The edge computing platform extracts features from the target image to determine its contour image;

[0165] The edge computing platform determines the target candidate region based on the contour image;

[0166] The edge computing platform sends target candidate regions and contour images to the cloud platform, which are used to perform target detection on the target image.

[0167] or

[0168] The cloud platform receives target candidate region and contour images sent by the edge computing platform;

[0169] The cloud platform performs target detection on the target image based on the target candidate region and contour image;

[0170] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0171] The target candidate region is determined by the edge computing platform based on the contour image.

[0172] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to execute the target detection methods provided in the above embodiments, such as including:

[0173] The edge computing platform extracts features from the target image to determine its contour image;

[0174] The edge computing platform determines the target candidate region based on the contour image;

[0175] The edge computing platform sends target candidate regions and contour images to the cloud platform, which are used to perform target detection on the target image.

[0176] or

[0177] The cloud platform receives target candidate region and contour images sent by the edge computing platform;

[0178] The cloud platform performs target detection on the target image based on the target candidate region and contour image;

[0179] The contour image is determined by the edge computing platform through feature extraction of the target image;

[0180] The target candidate region is determined by the edge computing platform based on the contour image.

[0181] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A target detection method, characterized in that, include: The edge computing platform extracts features from the target image to determine the contour image of the target image; The edge computing platform determines the target candidate region based on the contour image; The edge computing platform sends the target candidate region and the contour image to the cloud platform, and the target candidate region and the contour image are used to perform target detection on the target image; The edge computing platform performs feature extraction on the target image to determine the contour image of the target image, including: The edge computing platform extracts the contour features of the target image, and the edge computing platform determines the difference regions between the target image and its adjacent frames in the target video; If the area of ​​the difference region does not exceed an area threshold, the edge computing platform extracts the contour features of the difference region. The edge computing platform superimposes the contour features of the difference region onto the contour features of the target image, and determines the feature image of the superimposed target image based on the superimposed contour features. The edge computing platform divides the feature image into multiple target units of target pixel value size, and determines the contour image based on the multiple target units.

2. The target detection method according to claim 1, characterized in that, The edge computing platform determines the target candidate region based on the contour image, including: The edge computing platform determines the target color block based on the target unit in the contour image; The difference in grayscale values ​​between any two adjacent target units within the target color block is less than the grayscale value threshold. The edge computing platform determines the target candidate region based on the vertex coordinates of the target color block, as well as the length and width of the target color block.

3. The target detection method according to any one of claims 1-2, characterized in that, The edge computing platform sends the target candidate region and the contour image to the cloud platform, including: The edge computing platform determines the weight corresponding to the level of the access service based on the preset service level. The edge computing platform determines the bandwidth occupied by the access service level based on the weight. The edge computing platform sends the target candidate region and the contour image to the cloud platform based on the bandwidth.

4. A target detection method, characterized in that, include: The cloud platform receives target candidate region and contour images sent by the edge computing platform; The cloud platform performs target detection on the target image based on the target candidate region and the contour image; The contour image is determined by the edge computing platform through feature extraction of the target image; The target candidate region is determined by the edge computing platform based on the contour image; The contour image is determined based on the following method: The edge computing platform extracts the contour features of the target image, and the edge computing platform determines the difference regions between the target image and its adjacent frames in the target video; If the area of ​​the difference region does not exceed an area threshold, the edge computing platform extracts the contour features of the difference region. The edge computing platform superimposes the contour features of the difference region onto the contour features of the target image, and determines the feature image of the superimposed target image based on the superimposed contour features. The edge computing platform divides the feature image into multiple target units of target pixel value size, and determines the contour image based on the multiple target units.

5. A target detection device, characterized in that, include: The first determining module is used by the edge computing platform to extract features from the target image and determine the contour image of the target image; The second determining module is used by the edge computing platform to determine the target candidate region based on the contour image; The sending module is used by the edge computing platform to send the target candidate region and the contour image to the cloud platform, wherein the target candidate region and the contour image are used to perform target detection on the target image; The edge computing platform performs feature extraction on the target image to determine the contour image of the target image, including: The edge computing platform extracts the contour features of the target image, and the edge computing platform determines the difference regions between the target image and its adjacent frames in the target video; If the area of ​​the difference region does not exceed an area threshold, the edge computing platform extracts the contour features of the difference region. The edge computing platform superimposes the contour features of the difference region onto the contour features of the target image, and determines the feature image of the superimposed target image based on the superimposed contour features. The edge computing platform divides the feature image into multiple target units of target pixel value size, and determines the contour image based on the multiple target units.

6. A target detection device, characterized in that, include: The receiving module is used by the cloud platform to receive target candidate region and contour images sent by the edge computing platform; The detection module is used by the cloud platform to perform target detection on the target image based on the target candidate region and the contour image; The contour image is determined by the edge computing platform through feature extraction of the target image; The target candidate region is determined by the edge computing platform based on the contour image; The contour image is determined based on the following method: The edge computing platform extracts the contour features of the target image, and the edge computing platform determines the difference regions between the target image and its adjacent frames in the target video; If the area of ​​the difference region does not exceed an area threshold, the edge computing platform extracts the contour features of the difference region. The edge computing platform superimposes the contour features of the difference region onto the contour features of the target image, and determines the feature image of the superimposed target image based on the superimposed contour features. The edge computing platform divides the feature image into multiple target units of target pixel value size, and determines the contour image based on the multiple target units.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the target detection method according to any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the target detection method as described in any one of claims 1 to 4.

Images