Wind speed identification method, device, equipment and medium

Abstract

The invention discloses a wind speed recognition method and device, equipment and a medium, and relates to the technical field of computers, the method comprises the steps that an image frame sequence is acquired, and the image frame sequence comprises multiple frames of images which are continuous in time; performing optical flow feature extraction on the image frame sequence to obtain an optical flow feature representing a motion state of a target object in the image frame sequence; processing the optical flow feature from a time dimension and a space dimension to obtain a time dimension feature and a space dimension feature; and determining the wind speed matched with the image frame sequence according to the time dimension feature and the space dimension feature. The method can improve the accuracy of wind speed identification.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, device, equipment and medium for wind speed recognition. Background Art

[0002] With the rapid development of technology, as an immersive interactive experience space, the 8D cockpit has been widely used in fields such as entertainment and simulation training. In the 8D cockpit, wind speed simulation is one of the key factors in creating a realistic environment. Accurate wind speed recognition can enable users to truly feel the real wind feeling as if they were in a virtual scene, greatly enhancing the immersion and authenticity of the experience.

[0003] Currently, a convolutional neural network can be used to recognize the wind speed from a single image. However, a single image is easily affected by external environmental factors such as lighting and weather, resulting in poor robustness and generalization ability of the model and prone to overfitting problems.

[0004] For this reason, a technology for recognizing the wind speed from multiple images using the optical flow algorithm has been proposed. However, since the traditional optical flow algorithm is usually applicable to the motion analysis of a simplified background, when the background is complex or the contrast between the target object and the background is low, the optical flow algorithm is easily interfered, thereby affecting the recognition result and further reducing the accuracy of wind speed recognition. Summary of the Invention

[0005] This application provides a method, device, equipment and medium for wind speed recognition, which can improve the accuracy of wind speed recognition.

[0006] To achieve the above object, this application adopts the following technical solutions: In a first aspect, this application provides a method for wind speed recognition, and the method includes: Obtain an image frame sequence, where the image frame sequence includes multiple frames of images that are continuous in time; Extract optical flow features from the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; Process the optical flow features from the time dimension and the space dimension respectively to obtain time dimension features and space dimension features; Determine the wind speed matching the image frame sequence according to the time dimension features and the space dimension features.

[0007] Optionally, the determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features includes: Input the time dimension feature and the space dimension feature into a wind speed prediction model to obtain a wind speed that matches the image frame sequence; wherein, the wind speed prediction model is trained based on sample time dimension features, sample space dimension features, and sample labels, the sample time dimension features and the sample space dimension features are obtained based on sample optical flow features, and the sample optical flow features are obtained from the sample image frame sequence.

[0008] Optionally, the time dimension features include: average acceleration of multiple frames of images, rate of change of speed of multiple frames of images, and direction consistency of multiple frames of images; the space dimension features include average speed of multiple frames of images, speed gradient of multiple frames of images, speed variance of multiple frames of images, and proportion of average moving area of multiple frames of images; determining the wind speed that matches the image frame sequence according to the time dimension feature and the space dimension feature includes: Determine a time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images; determine a space dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of average moving area of multiple frames of images; Use the time dimension correction coefficient and the space dimension correction coefficient to correct the average speed of multiple frames of images to obtain a wind speed that matches the image frame sequence.

[0009] Optionally, determining the time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images includes:

[0010] Wherein, represents the time dimension correction coefficient, represents the average acceleration of multiple frames of images, represents the rate of change of speed of multiple frames of images, represents the direction consistency of multiple frames of images, represents the first sub-time weight, represents the second sub-time weight, represents the third sub-time weight.

[0011] Optionally, determining the space dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of average moving area of multiple frames of images includes:

[0012] Wherein, represents the space dimension correction coefficient, represents the speed gradient of multiple frames of images, represents the speed variance of multiple frames of images, Indicates the proportion of the average motion area of multiple frames of images, Indicates the weight of the first subspace, Indicates the weight of the second subspace, Indicates the weight of the third subspace.

[0013] Optionally, using the time dimension correction coefficient and the space dimension correction coefficient to correct the average speed of the multiple frames of images to obtain the wind speed matching the image frame sequence, including:

[0014]

[0015] Wherein, Indicates the time dimension correction coefficient, Indicates the space dimension correction coefficient, Indicates the wind speed matching the image frame sequence, Indicates the average speed of multiple frames of images, Indicates the time weight, and S indicates the space weight.

[0016] Optionally, the method further includes: Controlling the air volume of the air conditioner in the vehicle according to the wind speed matching the image frame sequence.

[0017] In a second aspect, the present application provides a wind speed identification device, the device includes: An acquisition module, configured to acquire an image frame sequence, where the image frame sequence includes multiple frames of images that are continuous in time; A feature extraction module, configured to perform optical flow feature extraction on the image frame sequence to obtain an optical flow feature representing the motion state of the target object in the image frame sequence; A spatio-temporal processing module, configured to process the optical flow feature from the time dimension and the space dimension respectively to obtain a time dimension feature and a space dimension feature; An identification module, configured to determine the wind speed matching the image frame sequence according to the time dimension feature and the space dimension feature.

[0018] In a third aspect, the present application provides a computing device, including a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is caused to execute the method according to any one of the first aspect.

[0019] Fourthly, the present application provides a computer-readable storage medium for storing a computer program for executing the method according to any one of the first aspect.

[0020] As can be seen from the above technical solutions, the present application has at least the following beneficial effects: The present application provides a wind speed identification method, which includes obtaining an image frame sequence, where the image frame sequence includes multiple frames of images that are continuous in time; extracting optical flow features from the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; processing the optical flow features from the time dimension and the space dimension respectively to obtain time dimension features and space dimension features; through the decoupling design of the time dimension and the space dimension, the present application can effectively separate the target motion and background interference; determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features; the time dimension features reflect the dynamic change law of the wind speed, and the space dimension features reflect the spatial consistency of the wind speed. When the contrast between the target and the background is reduced by 30%, the fusion mechanism of time and space can reduce the recognition error by 42% compared with the traditional method, verifying the effectiveness of feature complementarity. In summary, through the decoupling of spatio-temporal features and the fusion of multi-dimensional information, this method systematically solves the robustness problem of traditional optical flow algorithms in complex scenarios and provides an innovative solution for high-precision wind speed identification.

[0021] It should be understood that the description of technical features, technical solutions, beneficial effects or similar languages in the present application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of features or beneficial effects means that at least one embodiment includes specific technical features, technical solutions or beneficial effects. Therefore, the descriptions of technical features, technical solutions or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in this embodiment can be combined in any appropriate manner. Those skilled in the art will understand that an embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. Description of the Drawings

[0022] Figure 1 It is a flowchart of a wind speed identification method provided by an embodiment of the present application; Figure 2 It is a flowchart of a wind speed identification device provided by an embodiment of the present application; Figure 3 It is a schematic diagram of a computing device provided by an embodiment of the present application. Detailed Embodiments

[0023] The terms "first", "second", "third", etc. in the description and drawings of this application are used to distinguish different objects, rather than to limit a specific order.

[0024] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0025] Currently, the commonly used wind speed identification methods can be roughly divided into two categories: single-image recognition based on deep learning and multi-frame image sequence recognition based on optical flow algorithms.

[0026] The first method extracts features from a single frame of image through a convolutional neural network (CNN) for wind speed identification. Such methods usually require collecting and annotating a large amount of image data, standardizing and data augmenting the images, constructing a deep learning model using convolutional layers and fully connected layers, and selecting appropriate regression loss functions and optimizers. During model training, the hyperparameters are adjusted to improve the model performance and avoid overfitting. However, there are significant limitations in the wind speed identification using a single frame of image in practical applications. A single frame of image can only capture the static information at a specific moment and lacks the dynamic data support for the change of wind speed over time. The lack of image data in the spatial and temporal dimensions limits the extraction of effective features directly related to the wind speed from the image, making it difficult for the model to accurately identify the wind speed. A single frame of image is easily affected by external environmental factors such as illumination and weather, resulting in poor robustness and generalization ability of the model and prone to overfitting problems. This limits the reliability of the model in practical applications.

[0027] The second method uses traditional optical flow algorithms to identify the wind speed from multi-frame images. This method first extracts the motion vector field in the image sequence through traditional optical flow algorithms (such as Lucas-Kanade or dense optical flow), and then indirectly infers the wind speed by analyzing the change of the vector field. However, the performance of this method depends to a large extent on specific environmental conditions and visible reference objects, and has the following deficiencies. In the case of a complex background, large errors may occur in the optical flow calculation, and when there are few visible reference objects in the scene, the quality of the generated motion vector field will also decrease significantly, affecting the wind speed identification accuracy. Since traditional optical flow algorithms are usually applicable to the motion analysis of a simplified background, when the background is complex or the contrast between the target object and the background is low, the optical flow algorithm is vulnerable to interference, thus affecting the prediction accuracy.

[0028] In view of this, an embodiment of the present application provides a wind speed identification method, which can be executed by a processing device. The processing device can be a terminal or a server. The terminal includes, but is not limited to, a smart phone, a tablet computer, a laptop computer, a personal digital assistant, or a smart wearable device, etc. The server can be a cloud server, such as a central server in a central cloud computing cluster or an edge server in an edge cloud computing cluster. Of course, the server can also be a server in a local data center. A local data center refers to a data center directly controlled by a user. Exemplarily, the processing device can be a vehicle-mounted terminal. This method systematically solves the robustness problem of traditional optical flow algorithms in complex scenarios through the decoupling of spatio-temporal features and the fusion of multi-dimensional information, providing an innovative solution for high-precision wind speed identification.

[0029] For ease of understanding, the application scenario of the present application will be introduced first. In this application scenario, taking the processing device as a vehicle-mounted terminal as an example, a user can watch a video through the vehicle-mounted terminal (also referred to as a vehicle-mounted large screen) in the vehicle cockpit. In order to enable the user to have an immersive feeling during the video viewing in the cockpit, the processing device can control the vehicle's air conditioning system. For example, it can control the temperature inside the vehicle and the air volume, so as to match the scene in the video, enabling the user to feel the scene in the video even in the cockpit. For example, when an animation of a gentle breeze blowing on the face is played in the video, the processing device can control the space to make the air conditioner blow a gentle breeze on the face, so that the user can feel the feeling of a gentle breeze blowing on the face in the cockpit, greatly enhancing the immersion and authenticity of the experience and improving the user experience.

[0030] In order to make the technical solution of the present application clearer and easier to understand, the technical solution of the present application will be introduced below in conjunction with the accompanying drawings, as Figure 1 shown, this figure is a flowchart of a wind speed identification method provided by an embodiment of the present application.

[0031] This method includes: S101. The processing device acquires an image frame sequence.

[0032] The image frame sequence includes multiple frames of images that are continuous in time.

[0033] In some examples, the image frame sequence can be all or part of the image frames for processing and identifying the played video. Exemplarily, if the played video is a 2-second 60-frame animation, the processing device can use the 2-second 60-frame animation as the image frame sequence, or extract 30 frames from it, that is, use the 2-second 30-frame animation as the image frame sequence.

[0034] After the processing device acquires the image frame sequence, it is necessary to preprocess the image frame sequence, including standardization, size adjustment, etc., so as to adapt to the input dimension requirements of subsequent feature extraction.

[0035] S102. The processing device extracts optical flow features from the image frame sequence to obtain the motion state and optical flow features of the target object in the image frame sequence.

[0036] In some examples, the processing device can extract optical flow features from the image frame sequence through a Recurrent All Pairs Field Transforms (RAFT) model. The RAFT model can calculate the motion features of two consecutive frames of images. After the processing device preprocesses the image frame sequence, batch data can be generated through a dataloader (a key component for efficiently processing batch data in deep learning tasks), and the data dimension is [B, F, C, H, W], where B is the batch size, representing how much data is sent to the RAFT model at a time. For example, F = 30 represents 30 frames of images for each data, C = 2 is the number of channels, H = 224 represents the height of the image, and W = 224 represents the width of the image.

[0037] In some embodiments, the processing device can sequentially select multiple pairs of consecutive two-frame image frames from the image frame sequence as image pairs, and then sequentially input them into the RAFT model to obtain the optical flow features of each image pair. The optical flow features of each image pair represent the motion vectors of each pixel point in the image and are used to represent the motion information between the frames (e.g., horizontal motion information and vertical motion information).

[0038] S103. The processing device processes the optical flow features from the time dimension and the space dimension respectively to obtain the time dimension features and the space dimension features.

[0039] In this application, after obtaining the optical flow features, the processing device does not directly rely on the optical flow features for wind speed recognition, but performs multi-dimensional processing on the optical flow features, which can enhance the feature expression ability and improve the calculation efficiency.

[0040] In some embodiments, the processing device can use the TimeSformer model to process the above optical flow features. The TimeSformer model can capture spatio-temporal information in the video through the self-attention mechanism and can effectively process the optical flow features in the image frame sequence to extract the time dimension features and the space dimension features respectively.

[0041] When processing optical flow features, the TimeSformer model treats the input optical flow feature sequence as a series of tokens, and then calculates the correlation between these tokens through a multi-head self-attention mechanism. In the spatial dimension, the TimeSformer model focuses on the relationship between optical flow features at different positions in the same frame; in the temporal dimension, the TimeSformer model focuses on the relationship between optical flow features at the same or different positions between different frames.

[0042] Exemplarily, the TimeSformer model includes a spatial attention module and a temporal attention module. The spatial attention module learns spatial features through a self-attention mechanism within each frame; the temporal attention module captures feature dependencies in the temporal dimension through a self-attention mechanism between frames.

[0043] Among them, the dimension of the optical flow feature is [B, T, C, H, W], B is the batch size, T represents the number of image pairs, for example, T=29, F=30 represents 30 frames of images per data, C=2 is the number of channels, H=224 represents the height of the image, and W =224 represents the width of the image. The feature embedding layer divides each frame of the input sequence into several small patches (Chinese: blocks, usually refers to local areas of fixed size) and flattens them to form an input format suitable for the Transformer model. The temporal attention module and the spatial attention module use two Transformer submodules to process the temporal dimension and the spatial dimension respectively to obtain temporal dimension features and spatial dimension features, ensuring that effective temporal dependency information and spatial dependency information are extracted within the sequence and frame.

[0044] S104: The processing device determines a wind speed that matches the image frame sequence according to the time dimension characteristics and the space dimension characteristics.

[0045] After the processing device obtains the time dimension features and the space dimension features, it can determine the wind speed that matches the image frame sequence based on the time dimension features and the space dimension features.

[0046] In some examples, the processing device may input the time dimension features and the space dimension features into the wind speed prediction model to obtain the wind speed matching the image frame sequence. The following describes the training method of the wind speed prediction model: Exemplarily, the processing device can obtain a sequence of sample image frames and the corresponding sample labels of the sequence of sample image frames, then extract the optical flow features from the sequence of sample image frames to obtain sample optical flow features, and then perform multi-dimensional processing on the sample optical flow features from the time dimension and the spatial dimension to obtain sample time dimension features and sample spatial dimension features. The sample time dimension features, sample spatial dimension features, and sample labels form a set of training data. The processing device can obtain multiple sets of training data in a similar manner, such as 1000 sets of training data, and then use these training data for model training to obtain a wind speed prediction model. Specifically, the processing device can calculate the loss based on the prediction result of the wind speed prediction model and the sample label. For example, using the mean squared error loss function, the calculated loss is used for backpropagation to update the gradients of each layer of the model, thereby adjusting the weight parameters. After multiple rounds of iterative training, when the model gradually stabilizes and the prediction error converges within a reasonable range, the final model weights are saved. In the inference stage, the model parameters remain unchanged. At this time, the output result obtained by processing the input short video (a sequence of consecutive image frames) through the wind speed prediction model is the wind speed matching the image frame sequence. For example, [0.3, 0.2, 0.5…0.9]. The processing device can take the average value or the maximum value as the overall wind speed, or the obtained result is just the sequence of the above wind speed values.

[0047] In some other embodiments, the time dimension features include the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images. The spatial dimension features include the average speed of multiple frames of images, the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of the average motion area of multiple frames of images. The processing device can determine the time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images, and determine the spatial dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of the average motion area of multiple frames of images. Then, the processing device uses the time dimension correction coefficient and the spatial dimension correction coefficient to correct the average speed of multiple frames of images to obtain the wind speed matching the image frame sequence.

[0048] Among them, the determination method of the average acceleration of multiple frames of images is as follows:

[0049] Among them, represents the average acceleration of multiple frames of images, represents the speed value determined by the k-th image frame, represents the speed value determined by the (k + 1)-th image frame, represents the time difference between two adjacent image frames, represents the total number of image frames.

[0050] The method for determining the rate of change of the multi-frame image speed is as follows:

[0051] Wherein, represents the rate of change of the multi-frame image speed, represents the speed value determined by the k-th image frame, represents the speed value determined by the (k - M)-th image frame, that is, the speed value M frames before.

[0052] The method for determining the direction consistency of the multi-frame image is as follows: The processing device first calculates the mean value of the dominant directions of K image frames:

[0053] Wherein, represents the mean value of the dominant directions of K image frames, represents the dominant direction (in radians) of the k-th image frame, and then calculates the direction consistency of the multi-frame image:

[0054] Wherein, represents the direction consistency of the multi-frame image, represents the mean value of the dominant directions of K image frames.

[0055] The method for determining the speed gradient of the multi-frame image is as follows:

[0056] Wherein, represents the speed gradient of the multi-frame image, represents the speed gradient of the k-th image frame.

[0057] The method for determining the speed variance of the multi-frame image is as follows:

[0058] represents the speed variance of the multi-frame image, represents the speed variance of the k-th image frame.

[0059] The method for determining the proportion of the average motion area of the multi-frame image is as follows:

[0060] represents the proportion of the average motion area of the multi-frame image, represents the proportion of the motion area of the k-th image frame.

[0061] In some embodiments, the processing device can determine the time dimension correction coefficient in the following manner:

[0062] Among them, represents the time dimension correction coefficient, represents the average acceleration of multiple-frame images, represents the rate of change of the speed of multiple-frame images, represents the direction consistency of multiple-frame images, represents the first sub-time weight, represents the second sub-time weight, represents the third sub-time weight.

[0063] In some embodiments, the processing device can determine the spatial dimension correction coefficient through the following formula:

[0064] Among them, represents the spatial dimension correction coefficient, represents the speed gradient of multiple-frame images, represents the speed variance of multiple-frame images, represents the proportion of the average motion area of multiple-frame images, represents the first sub-space weight, represents the second sub-space weight, represents the third sub-space weight.

[0065] In some embodiments, the processing device can use the time dimension correction coefficient and the spatial dimension correction coefficient to correct the average speed of multiple-frame images in the following manner to obtain the wind speed matching the image frame sequence:

[0066]

[0067] Among them, represents the wind speed matching the image frame sequence, represents the average speed of multiple-frame images, represents the time weight, and S represents the spatial weight. Among them, the time weight and the spatial weight can be customarily set. In some examples, T = 0.7, S = 0.3, and in other examples, T = S = 0.5.

[0068] In some embodiments, after the processing device determines the wind speed matching the image frame sequence, it can control the air volume output of the air conditioner in the vehicle, thereby simulating the air volume in the virtual scene, enabling the user to have an immersive feeling and enhancing the immersion of the user's movie viewing.

[0069] In some embodiments, the processing device can preset the correspondence between the wind speed and white noise. After the processing device determines the wind speed, it can determine the level of white noise corresponding to the wind speed based on the above correspondence. Then, while controlling the air conditioner, the processing device controls the vehicle's audio and video system to play white noise, further enhancing the user's immersion in watching movies.

[0070] Based on the above description, embodiments of the present application provide a wind speed identification method. The method includes obtaining an image frame sequence, where the image frame sequence includes multiple frames of images that are continuous in time; extracting optical flow features from the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; processing the optical flow features from the time dimension and the space dimension respectively to obtain time dimension features and space dimension features; through the decoupled design of the time dimension and the space dimension in the present application, the target motion and background interference can be effectively separated; determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features; the time dimension features reflect the dynamic change law of the wind speed, and the space dimension features reflect the spatial consistency of the wind speed. When the contrast between the target and the background decreases by 30%, the fusion mechanism of time and space can reduce the recognition error by 42% compared with the traditional method, verifying the effectiveness of feature complementarity. In summary, this method systematically solves the robustness problem of traditional optical flow algorithms in complex scenarios through the decoupling of spatio-temporal features and the fusion of multi-dimensional information, providing an innovative solution for high-precision wind speed identification.

[0071] As described above in combination with Figure 1 The wind speed identification method provided by the embodiments of the present application has been introduced in detail. Next, the devices and equipment provided by the embodiments of the present application will be introduced with reference to the accompanying drawings.

[0072] As Figure 2 shown, this figure is a schematic diagram of a wind speed identification device provided by an embodiment of the present application. The device includes: An acquisition module 201, configured to acquire an image frame sequence, where the image frame sequence includes multiple frames of images that are continuous in time; A feature extraction module 202, configured to extract optical flow features from the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; A spatio-temporal processing module 203, configured to process the optical flow features from the time dimension and the space dimension respectively to obtain time dimension features and space dimension features; An identification module 204, configured to determine the wind speed matching the image frame sequence according to the time dimension features and the space dimension features.

[0073] Optionally, the recognition module 204 is specifically configured to input the time dimension feature and the space dimension feature into a wind speed prediction model to obtain a wind speed that matches the image frame sequence; wherein, the wind speed prediction model is trained based on sample time dimension features, sample space dimension features, and sample labels, the sample time dimension features and the sample space dimension features are obtained based on sample optical flow features, and the sample optical flow features are obtained from the sample image frame sequence.

[0074] Optionally, the time dimension features include: average acceleration of multiple frames of images, rate of change of speed of multiple frames of images, and direction consistency of multiple frames of images; the space dimension features include average speed of multiple frames of images, speed gradient of multiple frames of images, speed variance of multiple frames of images, and proportion of average moving area of multiple frames of images; the recognition module 204 is specifically configured to determine a time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images; determine a space dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of average moving area of multiple frames of images; and use the time dimension correction coefficient and the space dimension correction coefficient to correct the average speed of multiple frames of images to obtain a wind speed that matches the image frame sequence.

[0075] Optionally, the recognition module 204 is specifically configured to determine the time dimension correction coefficient in the following manner:

[0076] Wherein, represents the time dimension correction coefficient, represents the average acceleration of multiple frames of images, represents the rate of change of speed of multiple frames of images, represents the direction consistency of multiple frames of images, represents the first sub-time weight, represents the second sub-time weight, represents the third sub-time weight.

[0077] Optionally, the recognition module 204 is specifically configured to determine the space dimension correction coefficient in the following manner:

[0078] Wherein, represents the space dimension correction coefficient, represents the speed gradient of multiple frames of images, represents the speed variance of multiple frames of images, represents the proportion of average moving area of multiple frames of images, represents the first sub-space weight, represents the second sub-space weight, Represents the weight of the third subspace.

[0079] Optionally, the recognition module 204 is specifically configured to obtain the wind speed matching the image frame sequence in the following manner:

[0080]

[0081] Wherein, Represents the wind speed matching the image frame sequence, Represents the average speed of multiple frames of images, Represents the time weight, and S represents the space weight.

[0082] Optionally, the device further includes a control module, which is configured to control the air volume output of the air conditioner in the vehicle according to the wind speed matching the image frame sequence.

[0083] The wind speed recognition device according to the embodiment of the present application can correspond to executing the method described in the embodiment of the present application, and the above other operations and / or functions of each module / unit of the wind speed recognition device respectively aim to implement Figure 1 The corresponding processes of the respective methods in the illustrated embodiments, and for the sake of brevity, will not be described in detail here.

[0084] The embodiment of the present application further provides a computing device. As Figure 3 shown, this figure is a schematic diagram of a computing device provided by the embodiment of the present application. The computing device 300 includes a bus 301, a processor 302, a communication interface 303, and a memory 304. The processor 302, the memory 304, and the communication interface 303 communicate with each other through the bus 301.

[0085] The bus 301 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 3 only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0086] The processor 302 can be any one or more of processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0087] The communication interface 303 is used for external communication.

[0088] The memory 304 can include volatile memory, such as random access memory (RAM). The memory 304 can also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0089] Executable code is stored in the memory 304, and the processor 302 executes the executable code to perform the foregoing wind speed identification method.

[0090] Specifically, in the case of implementing Figure 2 the illustrated embodiment, and Figure 2 when each module or unit of the wind speed identification device described in the embodiment is implemented by software, the software or program code required to execute the functions of each module / unit in Figure 2 can be partially or entirely stored in the memory 304. The processor 302 executes the program code corresponding to each unit stored in the memory 304 to perform the foregoing wind speed identification method.

[0091] An embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store or a data storage device such as a data center that includes one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state drive). The computer-readable storage medium includes instructions that direct the computing device to execute the foregoing wind speed identification method.

[0092] An embodiment of the present application also provides a computer program product that includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, they generate, in whole or in part, the processes or functions described in the embodiments of the present application.

[0093] The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer or data center to another website, computer or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0094] When the computer program product is executed by a computer, the computer executes any of the aforementioned wind speed identification methods. The computer program product may be a software installation package, and when any of the aforementioned wind speed identification methods is needed, the computer program product may be downloaded and executed on a computer.

[0095] The descriptions of the processes or structures corresponding to the above-mentioned figures have different emphases. For parts that are not described in detail in a certain process or structure, please refer to the relevant descriptions of other processes or structures.

[0096] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be included in the protection scope of the present application.

Claims

1. A wind speed identification method, characterized in that, The method includes: Obtaining an image frame sequence, where the image frame sequence includes multiple frames of images that are temporally continuous; Performing optical flow feature extraction on the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; Processing the optical flow features respectively from the time dimension and the space dimension to obtain time dimension features and space dimension features; Determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features.

2. The method according to claim 1, wherein The determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features includes: Inputting the time dimension features and the space dimension features into a wind speed prediction model to obtain the wind speed matching the image frame sequence; wherein, the wind speed prediction model is trained based on sample time dimension features, sample space dimension features and sample labels, the sample time dimension features and the sample space dimension features are obtained based on sample optical flow features, and the sample optical flow features are obtained based on the sample image frame sequence.

3. The method according to claim 1, wherein The time dimension features include: average acceleration of multiple frames of images, rate of change of speed of multiple frames of images, and direction consistency of multiple frames of images; the space dimension features include average speed of multiple frames of images, speed gradient of multiple frames of images, speed variance of multiple frames of images, and proportion of average motion area of multiple frames of images; the determining the wind speed matching the image frame sequence according to the time dimension features and the space dimension features includes: Determining a time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images; determining a space dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of average motion area of multiple frames of images; Using the time dimension correction coefficient and the space dimension correction coefficient to correct the average speed of the multiple frames of images to obtain the wind speed matching the image frame sequence.

4. The method according to claim 3, characterized in that The determining a time dimension correction coefficient according to the average acceleration of multiple frames of images, the rate of change of speed of multiple frames of images, and the direction consistency of multiple frames of images includes: Among them, represents the time dimension correction coefficient, represents the average acceleration of multiple frames of images, represents the rate of change of velocity of multiple frames of images, represents the direction consistency of multiple frames of images, represents the first sub-time weight, represents the second sub-time weight, represents the third sub-time weight.

5. The method according to claim 3, characterized in that, The determining a space dimension correction coefficient according to the speed gradient of multiple frames of images, the speed variance of multiple frames of images, and the proportion of average motion area of multiple frames of images includes: Among them, represents the spatial dimension correction coefficient, represents the multi-frame image velocity gradient, represents the multi-frame image velocity variance, represents the proportion of the average motion area of the multi-frame image, represents the first subspace weight, represents the second subspace weight, represents the third subspace weight.

6. The method according to claim 3, characterized in that Using the time dimension correction coefficient and the space dimension correction coefficient to correct the average speed of the multiple frames of images to obtain the wind speed matching the image frame sequence includes: Among them, represents the time dimension correction coefficient, represents the space dimension correction coefficient, represents the wind speed matching the image frame sequence, represents the average speed of multiple frames of images, represents the time weight, and S represents the space weight.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Controlling the air volume output of the air conditioner in the vehicle according to the wind speed matching the image frame sequence.

8. An air velocity recognition device, characterized in that, The device includes: An obtaining module, configured to obtain an image frame sequence, where the image frame sequence includes multiple frames of images that are temporally continuous; A feature extraction module, configured to perform optical flow feature extraction on the image frame sequence to obtain optical flow features representing the motion state of the target object in the image frame sequence; A spatio-temporal processing module, configured to process the optical flow features respectively from the time dimension and the space dimension to obtain time dimension features and space dimension features; An identification module, configured to determine a wind speed matching the image frame sequence according to the time dimension feature and the space dimension feature.

9. A computing device, characterized in that, Comprising a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is caused to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images