High-speed object recognition method, device and equipment

Abstract

The application relates to the technical field of image processing, in particular to a high-speed object recognition method, device and equipment. The high-speed object recognition method comprises the following steps: collecting continuous original image frames of a target scene; determining a fast motion region image from the continuous original image frames through a double-flow network structure; and determining a high-speed object in the target scene in the fast motion region to highlight the high-speed object through weighting. According to the embodiment of the application, the spatial features and the motion features of the target scene around the vehicle are identified through the double-flow network structure, the high-speed object is distinguished from other static objects or slowly moving objects, the high-speed object is highlighted through the reaction mechanism of the motion sensitive neuron, and more attention is paid to the high-speed object, so that the high-speed object can be quickly recognized in many automatic driving scenes, and the safety of the automatic driving is ensured.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and specifically to a high-speed object recognition method, apparatus, and device. Background Technology

[0002] In current autonomous driving technologies, to ensure vehicle safety, autonomous driving systems must handle diverse road conditions and the detection and tracking of high-speed objects to avoid them. However, existing autonomous driving systems still face many challenges when tracking small, high-speed moving targets. These challenges include, but are not limited to, target recognition in dynamic backgrounds, maintaining high-precision tracking under complex lighting and weather conditions, and insufficient ability to quickly predict the future position of targets. This results in the vehicle's autonomous driving system failing to accurately identify high-speed objects, thus causing driving safety issues. Summary of the Invention

[0003] This invention provides a high-speed object recognition method, apparatus, and device to solve the problem that existing vehicle autonomous driving systems cannot accurately recognize high-speed objects, thus causing driving safety issues.

[0004] In a first aspect, embodiments of this aspect provide a high-speed object recognition method, the method comprising: Acquire continuous raw image frames of the target scene; The fast-moving region image is determined from the consecutive raw image frames using a two-stream network structure; High-speed objects in the target scene within the fast-moving region are identified and highlighted with weight.

[0005] Optionally, determining the fast-moving region image from the consecutive raw image frames using a two-stream network structure includes: The edge images corresponding to each consecutive original image frame are determined by an edge detection algorithm; Determine the difference image between each pair of consecutive original image frames that are adjacent to each other; The edge image and the difference image are fused to obtain a motion feature image; Spatial features are determined from the edge image through the spatial flow branch of the dual-flow network structure; Motion features are determined from the motion feature image through the temporal branch of the dual-stream network structure; The fast-moving region image is determined from the original image frame based on the spatial features and the motion features.

[0006] Optionally, fusing the edge image and the difference image to obtain the motion feature image includes: Based on a first weight pre-assigned to the edge image and a second weight pre-assigned to the difference image, the difference image and its corresponding edge image of each frame are fused into the motion feature image; or, A bitwise AND operation is performed on the pixel values ​​at corresponding positions in the difference images and corresponding edge images of each frame to fuse the difference images and their corresponding edge images into the motion feature image; or, Based on the channels of the differential images of each frame and the channels of the edge images corresponding to each frame's differential images, the differential images of each frame and their corresponding edge images are fused into a three-dimensional motion feature image.

[0007] Optionally, determining spatial features from the edge image through the spatial flow branch of the dual-stream network structure includes: Spatial features are extracted from the edge images of each single frame through the convolutional layers of the spatial flow branch; Complex spatial features are extracted from each single frame of the edge image by a first deep network composed of convolutional layers and pooling layers of the spatial flow branch. The extracted spatial features and complex spatial features are encoded into a vector by the average pooling layer of the spatial flow branch, which serves as the spatial features of the edge image. The process of determining motion features from the motion feature image through the temporal branch of the dual-stream network structure includes: Temporal features are extracted from each single-frame motion feature image through the convolutional layers of the temporal flow branch. The second deep network of the time-flow branch extracts complex motion features from each single frame motion feature image; The extracted motion features and complex motion features are encoded into a vector by the average pooling layer of the spatial flow branch, which serves as the motion features of the motion feature image.

[0008] Optionally, determining the fast-moving region image from the original image frame based on the spatial features and the motion features includes: The spatial features output by the spatial flow branch and the motion features output by the temporal flow branch are fused into target features; The dynamic changes of the original image frame are determined from the target features by the context-aware module; Based on whether the dynamic changes of the original image frame reach a threshold, the original image frame is divided into a static region and a moving region, wherein the threshold is determined by an adaptive thresholding method. The motion region is mapped to the image of the fast motion region.

[0009] Optionally, the step of identifying and highlighting high-speed objects in the target scene within the fast-moving region includes: The spatiotemporal convolutional layer of the neural network model is used to convolve multiple frames of the fast-moving region images to obtain spatiotemporal features from the fast-moving region images; The obtained spatiotemporal features are enhanced by the nonlinear activation function to highlight high-speed objects in the image of the fast-moving region.

[0010] Optionally, after obtaining the spatiotemporal features from the image of the fast-moving region, the method further includes: The obtained spatiotemporal features are normalized using a batch normalization layer in a neural network model.

[0011] Secondly, embodiments of the present invention provide a high-speed object recognition device, the device comprising: The acquisition module acquires continuous raw image frames of the target scene; The extraction module uses a dual-stream network structure to determine images of fast-moving regions from the continuous raw image frames; The marking module identifies high-speed objects in the target scene within the fast-moving region and highlights them with weights.

[0012] Thirdly, embodiments of the present invention provide an electronic device, including: At least one processor; and At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, which can invoke the program instructions to perform the method as described in any of the first aspects.

[0013] Fourthly, embodiments of the present invention provide a storage medium including a stored program, wherein, when the program is executed, it controls the device where the storage medium is located to perform the method described in any of the first aspects.

[0014] This invention simulates motion-sensitive neurons and uses a dual-stream network structure to identify the spatial and motion features of the target scene around the vehicle. It distinguishes high-speed objects from other static or slowly moving objects and highlights high-speed objects by simulating the reaction mechanism of motion-sensitive neurons, giving them more attention. This enables rapid identification of high-speed objects in many autonomous driving scenarios, ensuring the safety of autonomous driving. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The diagram shown is a flowchart of a high-speed object recognition method provided in an embodiment of this application; Figure 2 The diagram shown is a structural schematic of a high-speed object recognition device provided in an embodiment of this application; Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0018] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0019] This invention, through its embodiment, mimics the visual recognition and tracking methods of "motion-sensitive neurons" in animals such as frogs, designing a computational model that is particularly responsive to fast-moving objects. This model should be able to distinguish between static and slowly moving objects in an input image sequence, paying more attention to high-speed objects to optimize the dynamic object recognition and tracking capabilities of autonomous driving systems.

[0020] like Figure 1 The diagram shown is a flowchart of a high-speed object recognition method provided in an embodiment of the present invention. See also... Figure 1 The specific steps of this method include: S101, acquires continuous raw image frames of the target scene.

[0021] Specifically, during operation, the vehicle uses its onboard sensors to acquire images of the surrounding target scene and identifies high-speed objects within those images. To better determine the motion state of objects within the target scene, the vehicle captures multiple consecutive raw image frames during this process. After acquisition, the captured raw image frames undergo preprocessing, including brightness correction and color normalization for each frame. Noise reduction and scaling operations can also be performed on each frame if necessary.

[0022] Generally, the sensors carried by a vehicle can be cameras, meaning that the vehicle can use the cameras to capture continuous raw image frames of the target scene.

[0023] S102, the fast-moving region image is determined from consecutive raw image frames through a two-stream network structure.

[0024] Specifically, edge calculations are performed on each frame of the consecutive raw image frames output by S101 using an edge detection algorithm to highlight the object contours in each frame, thus obtaining the edge images corresponding to each consecutive raw image frame. Generally, edge calculations can be performed using edge detectors such as Sobel or Canny.

[0025] Determine the difference image between every two adjacent consecutive original image frames in the consecutive original image frames output by S101, and fuse the obtained difference image with its corresponding edge image to obtain a motion feature image containing edge information.

[0026] Specifically, when fusing the difference image and the edge image, it can be achieved through any of the following fusion methods: weighted fusion, logical operation fusion, and overlay layer fusion.

[0027] When performing weighted fusion using the `cv2.addWeighted` function, it is necessary to determine the first weight assigned to the edge image and the second weight assigned to the difference image. Based on the respective weights of the edge image and the difference image, each frame's difference image is fused with the corresponding edge image to form a motion feature image. Optionally, the first and second weights can be adjusted according to actual needs, but it must be ensured that the sum of the values ​​of the first and second weights is 1.

[0028] When performing logical operation fusion using the cv2.bitwise_and function, the pixel values ​​at corresponding positions in each frame difference image and the corresponding edge image are performed on-bit to fuse each frame difference image and its corresponding edge image into a motion feature image.

[0029] When performing overlay layer fusion using the np.dstack function, each frame difference image and edge image are treated as a channel. Based on the channels of each frame difference image and their corresponding edge images, the frame difference images and their corresponding edge images are fused into a three-dimensional motion feature image.

[0030] After preprocessing consecutive raw image frames, feature extraction is performed on the preprocessed consecutive raw image frames using a two-stream network structure.

[0031] Specifically, the two-stream network structure includes a spatial stream branch and a temporal stream branch. The spatial stream branch processes the edge image of a single frame, identifying object and scene features to extract spatial features. The temporal stream branch processes the inter-frame difference map fused with edge information, i.e., it processes the motion feature image, identifying changes and motion patterns to extract motion features.

[0032] In this process, when determining spatial features from edge images using the spatial flow branch of a two-stream network structure, spatial features are extracted from each single-frame edge image through convolutional layers of the spatial flow branch. Further, a first-depth network composed of convolutional and pooling layers from the spatial flow branch extracts complex spatial features from each single-frame edge image. After convolution in the final convolutional layer, the extracted spatial features and complex spatial features are encoded into a vector by the average pooling layer of the spatial flow branch, serving as the spatial features of the edge image. During feature extraction through convolutional layers, non-linearity can be increased using activation functions such as ReLU, and feature dimensionality can be reduced through pooling operations to improve the spatial invariance of the features.

[0033] When determining motion features from motion feature images using the temporal branch of a two-stream network structure, temporal features are extracted from each single-frame motion feature image through convolutional layers in the temporal branch. Further, complex motion features are extracted from each single-frame motion feature image through a second deep network in the temporal branch. Finally, the extracted motion features and complex motion features are encoded into a vector by the average pooling layer of the temporal branch, serving as the motion features of the motion feature image. During feature extraction through convolutional layers, activation functions and pooling layers can be used to increase non-linearity and reduce feature dimensionality. Furthermore, considering the continuity and temporal relationship of motion information, structures such as Recurrent Neural Networks (RNNs) or Long Short-Term Memory (LSTMs) are introduced as a second deep network to extract complex motion features.

[0034] After obtaining the output spatial features through the spatial flow branch of the two-stream network structure and the output motion features through the temporal flow branch of the two-stream network structure, the fast-moving region image is determined from the original image frame based on the spatial features and motion features.

[0035] Specifically, the spatial features output from the spatial flow branch and the motion features output from the temporal flow branch are fused into target features. A context-aware module determines the dynamic changes of the original image frame from the target features and checks whether these dynamic changes reach a threshold. Based on whether the dynamic changes of the original image frame reach the threshold, the original image frame is divided into static and moving regions, and the moving regions are mapped to generate fast-moving region images. The threshold is determined using an adaptive thresholding method.

[0036] The context-aware module can be an LSTM, used to understand dynamic changes in consecutive raw image frames based on target features and enhance motion detection.

[0037] S103 identifies high-speed objects in the target scene within the fast-moving region and highlights them with weighted emphasis.

[0038] Specifically, the multi-frame continuous images of fast-moving regions obtained through S102 are input into a pre-trained neural network model. Generally, the neural network model is a spatiotemporal convolutional network (STCN), which is highly responsive to rapidly changing features, can capture temporal changes in image sequences, and produces a strong response to rapid changes.

[0039] Spatiotemporal convolutional layers in a neural network model are used to convolve multiple frames of images of fast-moving regions to extract spatiotemporal features. Specifically, the spatiotemporal convolutional layers perform convolution operations on the input fast-moving images simultaneously in both temporal and spatial dimensions. Image regions at different time points are compared; regions that change rapidly, such as high-speed objects, receive a higher response, while regions that remain unchanged or change slowly, such as the background, receive a lower response, thus extracting the spatiotemporal features of the fast-moving regions.

[0040] By using a batch normalization layer in the neural network model to normalize the spatiotemporal features output by the convolution operation, the stability of training and the model's adaptability to different batches of data are enhanced, reducing potential problems such as gradient explosion.

[0041] The acquired spatiotemporal features are enhanced by nonlinear activation functions, such as ReLU or Leaky ReLU, to highlight high-speed objects in images of fast-moving regions. The final output is a weighted image in which high-speed objects stand out more compared to the background or slow-moving objects.

[0042] Furthermore, the network architecture, number of layers, kernel size, and other hyperparameters of the neural network model can be adjusted according to actual needs to optimize the model's performance in detecting fast motion. For example, the sensitivity of the neural network model to motion can be controlled by adjusting the kernel size and stride. Specifically, a larger kernel can be used to respond to more significant motion changes.

[0043] This invention simulates motion-sensitive neurons and uses a dual-stream network structure to identify the spatial and motion features of the target scene around the vehicle. It distinguishes high-speed objects from other static or slowly moving objects and highlights high-speed objects by simulating the reaction mechanism of motion-sensitive neurons, giving them more attention. This enables rapid identification of high-speed objects in many autonomous driving scenarios, ensuring the safety of autonomous driving.

[0044] Corresponding to the above-described high-speed object recognition method, this application also provides a high-speed object recognition device. See [link to relevant documentation]. Figure 2 This is a schematic diagram of the structure of a high-speed object recognition device provided in an embodiment of this application. The high-speed object recognition device may include: a data acquisition module 201, an extraction module 202, and a tagging module 203.

[0045] Acquisition module 201 acquires continuous raw image frames of the target scene; Extraction module 202 determines the fast motion region image from the consecutive raw image frames through a dual-stream network structure; The marking module 203 identifies high-speed objects in the target scene within the fast-moving region and highlights them with weight.

[0046] Figure 3 This is a schematic diagram illustrating the structure of one embodiment of the electronic device described in this specification. Figure 3 As shown, the above-mentioned electronic device may include at least one processor; and at least one memory communicatively connected to the processing unit, wherein: the memory stores program instructions executable by the processing unit, and the processor can execute the high-speed object recognition method provided in this embodiment by calling the program instructions.

[0047] The aforementioned electronic device can be a device capable of intelligent dialogue with the user, such as a cloud server. This specification does not limit the specific form of the electronic device in the embodiments. It is understood that the electronic device here refers to the machine mentioned in the method embodiments.

[0048] Figure 3 A block diagram of an exemplary electronic device suitable for implementing embodiments of this specification is shown. Figure 3The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments described in this specification.

[0049] like Figure 3 As shown, the electronic device is represented in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors 310, communication interface 320, memory 330, and communication bus 340 connecting different system components (including memory 330, communication interface 320 and processor 310).

[0050] Communication bus 340 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MAC) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0051] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0052] Memory 330 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 330 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments described herein.

[0053] A program / utility having a set (at least one) of program modules may be stored in memory 330. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of this specification.

[0054] The processor 310 executes various functional applications and data processing by running programs stored in the memory 330, such as implementing the high-speed object recognition method provided in the embodiments shown in this specification.

[0055] This specification provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute the high-speed object recognition method provided in the embodiments shown in this specification.

[0056] The aforementioned non-transitory computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0057] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0058] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0059] Computer program code for performing the operations described herein can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0060] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0061] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this specification, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0062] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this specification includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which the embodiments of this specification pertain.

[0063] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0064] It should be noted that the terminals involved in the embodiments of this specification may include, but are not limited to, personal computers (PCs), personal digital assistants (PDAs), wireless handheld devices, tablet computers, mobile phones, MP3 players, MP4 players, etc.

[0065] In the embodiments provided in this specification, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0066] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0067] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this specification.

[0068] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A high-speed object recognition method, characterized in that, The method includes: Acquire continuous raw image frames of the target scene; The fast-moving region image is determined from the consecutive raw image frames using a two-stream network structure; High-speed objects in the target scene within the fast-moving region are identified and highlighted with weight.

2. The method according to claim 1, characterized in that, The process of determining fast-moving region images from consecutive raw image frames using a two-stream network structure includes: The edge images corresponding to each consecutive original image frame are determined by an edge detection algorithm; Determine the difference image between each pair of consecutive original image frames that are adjacent to each other; The edge image and the difference image are fused to obtain a motion feature image; Spatial features are determined from the edge image through the spatial flow branch of the dual-flow network structure; Motion features are determined from the motion feature image through the temporal branch of the dual-stream network structure; The fast-moving region image is determined from the original image frame based on the spatial features and the motion features.

3. The method according to claim 2, characterized in that, The process of fusing the edge image and the difference image to obtain a motion feature image includes: Based on a first weight pre-assigned to the edge image and a second weight pre-assigned to the difference image, the difference image and its corresponding edge image of each frame are fused into the motion feature image; or, A bitwise AND operation is performed on the pixel values ​​at corresponding positions in the difference images and corresponding edge images of each frame to fuse the difference images and their corresponding edge images into the motion feature image; or, Based on the channels of the differential images of each frame and the channels of the edge images corresponding to each frame's differential images, the differential images of each frame and their corresponding edge images are fused into a three-dimensional motion feature image.

4. The method according to claim 2, characterized in that, The determination of spatial features from the edge image through the spatial flow branch of the dual-stream network structure includes: Spatial features are extracted from the edge images of each single frame through the convolutional layers of the spatial flow branch; Complex spatial features are extracted from each single frame of the edge image by a first deep network composed of convolutional layers and pooling layers of the spatial flow branch. The extracted spatial features and complex spatial features are encoded into a vector by the average pooling layer of the spatial flow branch, which serves as the spatial features of the edge image. The process of determining motion features from the motion feature image through the temporal branch of the dual-stream network structure includes: Temporal features are extracted from each single-frame motion feature image through the convolutional layers of the temporal flow branch. The second deep network of the time-flow branch extracts complex motion features from each single frame motion feature image; The extracted motion features and complex motion features are encoded into a vector by the average pooling layer of the spatial flow branch, which serves as the motion features of the motion feature image.

5. The method according to claim 2, characterized in that, Determining the fast-moving region image from the original image frame based on the spatial features and the motion features includes: The spatial features output by the spatial flow branch and the motion features output by the temporal flow branch are fused into target features; The dynamic changes of the original image frame are determined from the target features by the context-aware module; Based on whether the dynamic changes of the original image frame reach a threshold, the original image frame is divided into a static region and a moving region, wherein the threshold is determined by an adaptive thresholding method. The motion region is mapped to the image of the fast motion region.

6. The method according to claim 1, characterized in that, The step of identifying and highlighting high-speed objects in the target scene within the fast-moving region includes: The spatiotemporal convolutional layer of the neural network model is used to convolve multiple frames of the fast-moving region images to obtain spatiotemporal features from the fast-moving region images; The obtained spatiotemporal features are enhanced by the nonlinear activation function to highlight high-speed objects in the image of the fast-moving region.

7. The method according to claim 6, characterized in that, After obtaining the spatiotemporal features from the image of the fast-moving region, the method further includes: The obtained spatiotemporal features are normalized using a batch normalization layer in a neural network model.

8. A high-speed object recognition device, characterized in that, The device includes: The acquisition module acquires continuous raw image frames of the target scene; The extraction module uses a dual-stream network structure to determine images of fast-moving regions from the continuous raw image frames; The marking module identifies high-speed objects in the target scene within the fast-moving region and highlights them with weights.

9. An electronic device, characterized in that, include: At least one processor; as well as At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor can invoke the program instructions to perform the method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 7.

Images