An obstacle recognition method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide an obstacle identification method and device, electronic equipment and storage medium, and relate to the technical field of image processing. The method comprises receiving an image to be identified acquired by a camera; adding rain and snow noise to the image to be identified until the image becomes a pure noise image; denoising the pure noise image to obtain a rain and snow noise removed image; and performing obstacle detection on the rain and snow noise removed image to obtain a detection result. The method can accurately detect obstacles in rainy and snowy weather, has low requirements on imaging equipment, and solves the problem of obstacle detection in rainy and snowy weather under the premise of low cost and high efficiency.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to an obstacle recognition method, apparatus, electronic device, and storage medium. Background Technology

[0002] Currently, steel companies are pushing to create an all-weather, full-process, efficient and safe intelligent molten iron transportation system to realize intelligent molten iron dispatching and unmanned locomotive driving in the molten iron operation area, improve the accuracy and timeliness of molten iron transportation dispatching plans, and improve the operational efficiency of molten iron dispatching and locomotive dispatching.

[0003] To implement this system, full automation of the entire operation process is required. However, in steel mill environments, automated locomotives struggle to accurately detect obstacles in adverse weather conditions such as rain and snow, potentially leading to unsafe production operations. In the field of autonomous driving, existing obstacle detection methods, such as camera-based, millimeter-wave radar-based, and lidar-based technologies, are significantly affected by rain and snow, leading to frequent misjudgments of obstacles and high costs. Summary of the Invention

[0004] The purpose of this application is to provide an obstacle recognition method, device, electronic device, and storage medium that can accurately detect obstacles in rainy or snowy weather, with low requirements for imaging equipment, thus solving the problem of obstacle detection in rainy or snowy weather under the premise of low cost and high efficiency.

[0005] This application provides an obstacle recognition method, the method comprising:

[0006] Receive the image to be recognized from the camera;

[0007] Rain and snow noise is added to the image to be identified until it becomes a pure noise image;

[0008] The pure noise image is denoised to obtain a rain and snow noise-free image;

[0009] Obstacle detection is performed on the rain and snow noise-removed image to obtain the detection results.

[0010] In the above implementation process, the diffusion model principle is adopted to add rain and snow noise to the image to be identified acquired by the camera, and then perform rain and snow noise removal to obtain an image without rain and snow noise. This removes the influence of weather factors such as rain and snow on obstacle detection, thereby improving the detection accuracy. It also has low requirements for imaging equipment and low cost, solving the problem that existing methods are greatly affected by rain and snow weather, which makes it easy to misjudge obstacles and has high cost.

[0011] Further, adding rain and snow noise to the image to be identified until it becomes a pure noise image includes:

[0012] Add rain and snow noise to the image to be identified in the current state and sample it to obtain the sampled image for the current state;

[0013] Add rain and snow noise to the sampled image of the current state and sample it. Use the sampled image of the next state as the sampled image. Repeat the process of adding rain and snow noise and sampling until a pure noise image is obtained.

[0014] In the above implementation process, rain and snow noise is continuously added to the image to be identified, thereby turning the image to be identified into a pure noise image, so as to obtain a noise-free image based on the pure noise image.

[0015] Further, the denoising of the pure noise image to obtain a rain and snow noise-free image includes:

[0016] Construct a new Gaussian distribution, wherein the variance of the new Gaussian distribution is consistent with the variance of the Gaussian distribution of the sampled image, and initialize the mean of the new Gaussian distribution:

[0017] The process of narrowing the gap between the Gaussian distribution of the sampled image and the new Gaussian distribution is transformed into an optimization objective function.

[0018] The objective function is transformed by converting the Gaussian distribution of the sampled image and the mean of the new Gaussian distribution;

[0019] The model is updated through iterative training and gradient descent until it converges, resulting in a rain and snow noise-free image.

[0020] In the above implementation process, the pure noise image is filtered to obtain a noise-free image, thereby improving the accuracy of obstacle detection.

[0021] Further, the obstacle detection process performed on the de-raining and de-snowing noise image to obtain detection results includes:

[0022] The rain and snow noise-removed images are aggregated at different image fine-grained levels to generate a convolutional neural network for image features;

[0023] The image features are combined based on the convolutional neural network and then passed to the prediction layer to obtain the prediction result.

[0024] In the above implementation process, obstacle recognition can be performed on the image after removing rain and snow noise, and the recognition result can be accurately obtained.

[0025] This application embodiment also provides an obstacle recognition device, the device comprising:

[0026] The image receiving module is used to receive the image to be recognized acquired by the camera;

[0027] A rain and snow noise addition module is used to add rain and snow noise to the image to be identified until it becomes a pure noise image;

[0028] The denoising module is used to denoise the pure noise image to obtain a rain and snow noise-free image;

[0029] An obstacle detection module is used to detect obstacles in the de-raining and de-snowing noise image and obtain detection results.

[0030] In the above implementation process, the diffusion model principle is adopted to add rain and snow noise to the image to be identified acquired by the camera, and then perform rain and snow noise removal to obtain an image without rain and snow noise. This removes the influence of weather factors such as rain and snow on obstacle detection, thereby improving the detection accuracy. It also has low requirements for imaging equipment and low cost, solving the problem that existing methods are greatly affected by rain and snow weather, which makes it easy to misjudge obstacles and has high cost.

[0031] Furthermore, the rain and snow noise reduction module includes:

[0032] The noise-adding module is used to add rain and snow noise to the image to be identified in the current state and to sample it as the sampled image of the current state;

[0033] The sampling module is used to add rain and snow noise to the sampled image of the current state, and then sample it. The sampling result is used as the sampled image of the next state. The process of adding rain and snow noise and sampling is repeated until a pure noise image is obtained.

[0034] In the above implementation process, rain and snow noise is continuously added to the image to be identified, thereby turning the image to be identified into a pure noise image, so as to obtain a noise-free image based on the pure noise image.

[0035] Furthermore, the noise reduction module includes:

[0036] A new Gaussian distribution construction module is used to construct a new Gaussian distribution, wherein the variance of the new Gaussian distribution is consistent with the variance of the Gaussian distribution of the sampled image, and the mean of the new Gaussian distribution is initialized.

[0037] The objective function determination module is used to transform the process of narrowing the gap between the Gaussian distribution of the sampled image and the new Gaussian distribution into an optimization objective function;

[0038] The objective function transformation module is used to transform the objective function by transforming the Gaussian distribution of the sampled image and the mean of the new Gaussian distribution;

[0039] The rain and snow noise removal image acquisition module is used to update the model through iterative training and gradient descent until the model converges, thereby obtaining the rain and snow noise removal image.

[0040] In the above implementation process, the pure noise image is filtered to obtain a noise-free image, thereby improving the accuracy of obstacle detection.

[0041] Furthermore, the obstacle detection module includes:

[0042] The backbone network module is used to aggregate the rain and snow noise removal images at different image fine-grained levels to generate a convolutional neural network for image features;

[0043] The network layer module is used to combine image features based on the convolutional neural network and pass the image features to the prediction layer to obtain the prediction result.

[0044] In the above implementation process, obstacle recognition can be performed on the image after removing rain and snow noise, and the recognition result can be accurately obtained.

[0045] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the obstacle recognition method described in any one of the above-described embodiments.

[0046] This application also provides a readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the obstacle recognition method described in any one of the above-described embodiments. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating an obstacle recognition method provided in this application embodiment;

[0049] Figure 2 A flowchart of obstacle recognition in multi-level rain and snow weather based on Diffusion Model provided for embodiments of this application;

[0050] Figure 3 A flowchart for adding rain and snow noise provided in an embodiment of this application;

[0051] Figure 4 A flowchart for adding noise is provided for an embodiment of this application;

[0052] Figure 5 This is a schematic diagram of noise addition provided in an embodiment of this application;

[0053] Figure 6 A flowchart for removing rain and snow noise provided in the embodiments of this application;

[0054] Figure 7 This is a schematic diagram of rain and snow noise removal provided in an embodiment of this application;

[0055] Figure 8 This is a flowchart of obstacle detection provided in an embodiment of the present application;

[0056] Figure 9 A structural block diagram of an obstacle recognition device provided in an embodiment of this application;

[0057] Figure 10 This is a structural block diagram of another obstacle recognition device provided in an embodiment of this application.

[0058] icon:

[0059] 100 - Image receiving module; 200 - Rain / snow noise addition module; 201 - Noise addition module; 202 - Sampling module; 300 - Noise removal module; 301 - New Gaussian distribution construction module; 302 - Objective function determination module; 303 - Objective function transformation module; 304 - Rain / snow noise removal image acquisition module; 400 - Obstacle detection module; 401 - Backbone network module; 402 - Network layer module. Detailed Implementation

[0060] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0061] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0062] Example 1

[0063] Please refer to Figure 1 , Figure 1 This is a flowchart of an obstacle recognition method provided in an embodiment of this application.

[0064] This application utilizes the Diffusion Model principle, driven by the law of entropy increase. First, rain and snow noise is added to the image acquired by the camera, making it extremely chaotic. Then, a multi-level obstacle detection model based on the Diffusion Model can restore the order of the noisy image. Through continuous iterative training, an obstacle detection model that can eliminate rain and snow from images can be obtained, improving the accuracy of obstacle recognition. Figure 2 The diagram shown is a flowchart of obstacle recognition in multi-level rain and snow weather based on the Diffusion Model. It consists of two parts: rain and snow removal and obstacle detection.

[0065] The method specifically includes the following steps:

[0066] Step S100: Receive the image to be recognized acquired by the camera;

[0067] Step S200: Add rain and snow noise to the image to be identified until it becomes a pure noise image;

[0068] Step S300: Denoise the pure noise image to obtain a rain and snow noise-free image;

[0069] Step S400: Perform obstacle detection on the rain and snow noise removal image to obtain the detection results.

[0070] Specifically, such as Figure 3 The diagram shows a flowchart for adding rain and snow noise. Step S200 specifically includes the following steps:

[0071] Step S201: Add rain and snow noise to the image to be identified in the current state and sample it to obtain the sampled image for the current state;

[0072] Step S202: Add rain and snow noise to the sampled image of the current state and sample it. Use the sampling result as the sampled image of the next state. Repeat the process of adding rain and snow noise and sampling until a pure noise image is obtained.

[0073] Adding rain and snow noise is the forward pass process, which continuously adds rain and snow noise to the input data, i.e., the image to be detected, until it becomes pure noise (noise is the deviation between the true label and the actual label in the dataset). Based on prior knowledge, rain and snow noise follows a Gaussian distribution; by adding Gaussian noise to simulate rain and snow, the model converges faster.

[0074] like Figure 4 The diagram shows a flowchart for adding noise. Gaussian noise is added at each time step, and the image at the next time step is obtained from the image at the previous time step after adding Gaussian noise.

[0075] like Figure 5The diagram illustrates the addition of noise. The leftmost x0 represents the image to be detected, such as the dog image in the diagram; x t This represents a pure Gaussian noise image, such as the noise image on the far right of the corresponding figure; x t This represents x0 at time t with added noise, such as an image of a dog with added noise in the middle; q(x t |x t-1 Then it represents the previous state x. t-1 x follows a Gaussian distribution with mean x t It is sampled from the Gaussian distribution of the previous state.

[0076] The formula for adding noise is:

[0077]

[0078] in, β represents the mean. t I represents from β t It is obtained by sampling from a Gaussian distribution with variance.

[0079] The forward diffusion process can be understood as a Markov chain, which involves gradually adding Gaussian noise to a real image until it eventually becomes a pure Gaussian noise image.

[0080] like Figure 6 The flowchart shown is for removing rain and snow noise. Step S300 specifically includes the following steps:

[0081] Step S301: Construct a new Gaussian distribution, wherein the variance of the new Gaussian distribution is consistent with the variance of the Gaussian distribution of the sampled image, and initialize the mean of the new Gaussian distribution:

[0082] Step S302: Transform the process of narrowing the gap between the Gaussian distribution of the sampled image and the new Gaussian distribution into an optimization objective function;

[0083] Step S303: Transform the objective function by converting the mean of the Gaussian distribution of the sampled image and the new Gaussian distribution;

[0084] Step S304: Update the model through iterative training and gradient descent until the model converges to obtain the rain and snow noise-free image.

[0085] Step S300 is the reverse process (removing rain and snow noise), starting from random rain and snow noise and gradually restoring it to the original image without rain and snow noise – the denoising process.

[0086] like Figure 7 The diagram shown illustrates the process of removing rain and snow noise. The reverse diffusion process q(x) t-1 |x t ,x0) is the forward diffusion process q(xt |x t-1 The posterior probability distribution of q is obtained by progressively sampling from the rightmost pure Gaussian noise map (the top row in the diagram). In contrast to the forward process, x0 is obtained by the reverse process, sampling from the rightmost pure Gaussian noise map. When actually generating the image through the reverse process, x0 is unknown because it is the target image to be generated. By constructing a new Gaussian distribution p, the goal becomes reducing the gap between distributions p and q.

[0087] By continuously modifying the parameters of p to narrow the gap, p can replace q when they are sufficiently similar. This is because the posterior probability distribution q(x)... t-1 |x t Since x0 cannot be solved directly, a Gaussian distribution p(x0) is constructed. t-1 |x t (Second row in the figure), let its variance and posterior probability distribution q(x) t-1 |x t (x0) is consistent:

[0088] Its mean is set as:

[0089] Where, α t 1-β t , for This indicates that the noisy image obtained during the forward process changes with time t to x. t .

[0090] and q(x) t-1 |x t The difference between ,x0) is that x0 is changed to x θ (x t The expression ,t), is predicted by a backward process of a diffusion model, where the input to the backward diffusion model is the noisy image x. t And time step t. Then reduce the probability distribution p(x) t-1 |x t ) and q(x t-1 |x t The difference between x and x0 is transformed into optimizing the following objective function:

[0091]

[0092] However, if the model is allowed to directly draw from x... t Predicting x0 is too difficult to fit, given the preceding forward process where x is known. t We can get from x0: Transform the above formula as follows:

[0093]

[0094] Where ε represents rain and snow noise.

[0095] Substitute q(x) t-1 |x t From the mean of x0, we can obtain:

[0096]

[0097] From this formula, we can see that the posterior probability q(x) t-1 |x t The mean of x0 is only with x t This is related to the noise added at time step t during forward diffusion. Therefore, the constructed probability distribution p(x) can be similarly applied. t-1 |x t The mean of () is modified as follows:

[0098]

[0099] in, Represented as x t Gaussian noise added at time t.

[0100] The model is modified to predict the Gaussian noise ε added at the forward time step t, with the model input being x. t and time step t: The objective function for optimization then becomes:

[0101]

[0102] Then, through iterative training, each iteration first takes a real image x0 (the original input image data) from the dataset and samples a time step t from a uniform distribution;

[0103] Then, the noise ε is sampled from the standard Gaussian distribution, and x is calculated according to the formula. t .

[0104] The standard Gaussian distribution X ~ N(μ, σ^2) is a normal distribution with a mean of 0 and a standard deviation of 1, denoted as N(0, 1).

[0105] Next, x t The inputs t are fed into the Diffusion backward model, which outputs the predicted noise ε to fit the model. The model is then updated using gradient descent, and this process is repeated until the model converges.

[0106] like Figure 8The diagram shows the obstacle detection flowchart. Step S400 specifically includes the following steps:

[0107] Step S401: Aggregate the rain and snow noise removal images at different image fine-grained levels to generate a convolutional neural network for image features;

[0108] Step S402: Combine image features based on the convolutional neural network and pass the image features to the prediction layer to obtain the prediction result.

[0109] By performing forward and backward processes, an image with rain and snow noise removed can be obtained. Then, by feeding the image into the secondary detection model, a more accurate obstacle detection result can be obtained.

[0110] For example, a Yolov5 model can be used, which consists of four parts: input, backone, neck, and prediction.

[0111] The input is the image obtained after removing rain and snow noise (the image without rain and snow noise); Backbone: a convolutional neural network that aggregates and forms image features at different fine-grained levels; Neck: a series of network layers that mix and combine image features and pass the image features to the prediction layer; Head: predicts the image features, generates bounding boxes and predicts the category.

[0112] This method constructs a multi-level obstacle detection system. It uses a Diffusion model to filter out noise such as rain and snow in the image to obtain a rain- and snow-free image. During the forward pass of the Diffusion model, Gaussian noise is added to simulate rain and snow, which accelerates the convergence speed of the model. This method removes the influence of rain and snow on obstacle detection, thereby improving the accuracy of the detection results and reducing costs.

[0113] Example 2

[0114] This application provides an obstacle recognition device, applied to the obstacle recognition method described in Embodiment 1, such as... Figure 9 The diagram shown is a structural block diagram of an obstacle recognition device, which includes, but is not limited to:

[0115] Image receiving module 100 is used to receive the image to be recognized acquired by the camera;

[0116] Rain and snow noise module 200 is used to add rain and snow noise to the image to be identified until it becomes a pure noise image;

[0117] The denoising module 300 is used to denoise the pure noise image to obtain a rain and snow noise-free image;

[0118] The obstacle detection module 400 is used to perform obstacle detection on the rain and snow noise removal image and obtain detection results.

[0119] like Figure 10 The diagram shown is a structural block diagram of another obstacle recognition device, wherein the rain and snow noise module 200 includes:

[0120] The noise-adding module 201 is used to add rain and snow noise to the image to be recognized in the current state and sample it as the sampled image of the current state;

[0121] The sampling module 202 is used to add rain and snow noise to the sampled image of the current state, and then sample it. The sampling result is used as the sampled image of the next state. The process of adding rain and snow noise and sampling is repeated until a pure noise image is obtained.

[0122] The noise reduction module 300 includes:

[0123] The new Gaussian distribution construction module 301 is used to construct a new Gaussian distribution, wherein the variance of the new Gaussian distribution is consistent with the variance of the Gaussian distribution of the sampled image, and the mean of the new Gaussian distribution is initialized.

[0124] The objective function determination module 302 is used to transform the reduction of the gap between the Gaussian distribution of the sampled image and the new Gaussian distribution into an optimization objective function;

[0125] The objective function transformation module 303 is used to transform the objective function by transforming the mean of the Gaussian distribution of the sampled image and the new Gaussian distribution;

[0126] The rain and snow noise removal image acquisition module 304 is used to update the model through iterative training and gradient descent until the model converges, thereby obtaining the rain and snow noise removal image.

[0127] The obstacle detection module 400 includes:

[0128] The backbone network module 401 is used to aggregate the rain and snow noise removal images at different image fine-grained levels to generate a convolutional neural network for image features.

[0129] Network layer module 402 is used to combine image features based on the convolutional neural network and pass the image features to the prediction layer to obtain the prediction result.

[0130] This device uses the diffusion model principle to add rain and snow noise to the image to be identified acquired by the camera, and then performs rain and snow noise reduction to obtain an image without rain and snow noise. This removes the influence of weather factors such as rain and snow on obstacle detection, thereby improving the detection accuracy. It also has low requirements for imaging equipment and low cost, solving the problem that existing methods are greatly affected by rain and snow weather, which makes it easy to misjudge obstacles and has high costs.

[0131] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the obstacle recognition method described in Embodiment 1.

[0132] This application also provides a readable storage medium, characterized in that the readable storage medium stores computer program instructions, which are read and executed by a processor to perform the obstacle recognition method described in Embodiment 1.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0134] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0135] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they 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 portion 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.

[0136] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0137] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0138] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. An obstacle recognition method, characterized in that, The method includes: Receive the image to be recognized from the camera; Adding rain and snow noise to the image to be identified until it becomes a pure noise image specifically includes: adding rain and snow noise to the image to be identified in the current state and sampling it to obtain the sampled image of the current state; adding rain and snow noise to the sampled image of the current state and sampling it, using the sampling result as the sampled image of the next state, and repeating the adding rain and snow noise and sampling actions until a pure noise image is obtained. Denoising the pure noise image to obtain a rain and snow-free noise-free image specifically includes: constructing a new Gaussian distribution, wherein the new Gaussian distribution... p ( x t-1 | x t The variance of the sampled image and the Gaussian distribution q ( x t-1 | x t , x The variances of the sampled image Gaussian distribution and the new Gaussian distribution are consistent, and the mean of the new Gaussian distribution is initialized: reducing the difference between the sampled image Gaussian distribution and the new Gaussian distribution is transformed into an optimization objective function; by transforming the sampled image Gaussian distribution... q ( x t-1 | x t , x 0) and the new Gaussian distribution p ( x t-1 | x t The mean of the sampled images is used to transform the objective function, specifically including: x t and the image to be identified x The transformation relationship of 0 is expressed as: ,in, Indicates rain and snow noise. For 1- , for , The variance of the Gaussian distribution when adding rain and snow noise is the Gaussian distribution of the sampled image. q ( x t-1 | x t , x The mean transformation of 0) is: The new Gaussian distribution p ( x t-1 | x t The mean transformation of () is: , Represented as x t exist t Gaussian noise is added at each step; it is updated through iterative training and gradient descent until the model converges, resulting in a rain and snow noise-free image. Obstacle detection is performed on the rain and snow noise-removed image to obtain the detection results.

2. The obstacle recognition method according to claim 1, characterized in that, The obstacle detection process for the de-raining and de-snowing noise image, to obtain detection results, includes: The rain and snow noise-removed images are aggregated at different image fine-grained levels to generate a convolutional neural network for image features; The image features are combined based on the convolutional neural network and then passed to the prediction layer to obtain the prediction result.

3. An obstacle recognition device, characterized in that, The device includes: The image receiving module is used to receive the image to be recognized acquired by the camera; The rain and snow noise addition module is used to add rain and snow noise to the image to be identified until it becomes a pure noise image. Specifically, it includes: a noise addition module, which adds rain and snow noise to the image to be identified in the current state and samples it as the sampled image of the current state; a sampling module, which adds rain and snow noise to the sampled image of the current state and samples it, and uses the sampling result as the sampled image of the next state. The rain and snow noise addition and sampling actions are repeated until a pure noise image is obtained. The denoising module, used to denoise the pure noise image to obtain a rain and snow noise-free image, specifically includes: a new Gaussian distribution construction module, used to construct a new Gaussian distribution, wherein the new Gaussian distribution... p ( x t-1 | x t The variance of the sampled image and the Gaussian distribution q ( x t-1 | x t , x The variances of the sampled image Gaussian distribution and the new Gaussian distribution are consistent, and the mean of the new Gaussian distribution is initialized. The objective function determination module is used to transform reducing the difference between the sampled image Gaussian distribution and the new Gaussian distribution into an optimization objective function. The objective function transformation module is used to transform the sampled image Gaussian distribution... q ( x t-1 | x t , x 0) and the new Gaussian distribution p ( x t-1 | x t The mean of the sampled images is used as the objective function for transformation. x t and the image to be identified x The transformation relationship of 0 is expressed as: ,in, Indicates rain and snow noise. For 1- , for , The variance of the Gaussian distribution when adding rain and snow noise is the Gaussian distribution of the sampled image. q ( x t-1 | x t , x The mean transformation of 0) is: The new Gaussian distribution p ( x t-1 | x t The mean transformation of () is: , Represented as x t exist t Gaussian noise added at each step; the rain and snow noise removal image acquisition module is used to update the image through iterative training and gradient descent until the model converges, thereby obtaining the rain and snow noise removal image; An obstacle detection module is used to detect obstacles in the de-raining and de-snowing noise image and obtain detection results.

4. The obstacle recognition device according to claim 3, characterized in that, The obstacle detection module includes: The backbone network module is used to aggregate the rain and snow noise removal images at different image fine-grained levels to generate a convolutional neural network for image features; The network layer module is used to combine image features based on the convolutional neural network and pass the image features to the prediction layer to obtain the prediction result.

5. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the obstacle recognition method according to any one of claims 1 to 2.

6. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions, which, when read and executed by a processor, perform the obstacle recognition method according to any one of claims 1 to 2.

Images