Road safety inspection scene assessment simulation system and method based on image processing

Abstract

The invention discloses a road safety inspection scene assessment simulation system and method based on image processing. The system comprises an acquisition module, an extraction module, a construction module and an assessment module. The acquisition module is used for acquiring a scene image of a road safety inspection scene and preprocessing the scene image to obtain image data; the extraction module is used for extracting scene features based on the image data; the construction module is used for constructing a recognition model based on the scene features; and the assessment module is used for completing assessment simulation of the road safety inspection scene by using the identification model. Interference of human factors is reduced by using an automation technology, and diversity and adaptability of a simulation scene are improved through a machine learning algorithm, so that an assessment environment close to an actual combat is provided for law enforcement officers and workers, and manpower, material resources and time investment required by actual scene simulation is reduced.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a road safety inspection scene assessment simulation system and method based on image processing. Background Art

[0002] In the current field of road traffic safety management, assessment simulation, as a key means to improve the professional skills of law enforcement officers and staff, is facing challenges such as limitations of the simulation environment, subjectivity of assessment standards, resource cost constraints, and backward technical means. Traditional assessment simulation methods rely on manually set scenarios, which are difficult to fully cover the complex situations in actual work, and the assessment results are easily affected by personal experience and subjective judgment. There is a lack of unified assessment standards and automated evaluation tools. In addition, actual scenario simulation requires a lot of manpower, material resources and time investment, which limits the popularity and frequency of assessment simulation. Summary of the invention

[0003] In order to solve the technical problems in the above background, the present invention proposes an innovative assessment simulation system and method, which aims to achieve a more efficient, objective and comprehensive assessment simulation environment, thereby effectively improving the professional skills of road safety inspectors, strengthening road traffic safety management, and reducing the occurrence of safety accidents.

[0004] To achieve the above-mentioned purpose, the present invention provides a road safety inspection scene assessment simulation system based on image processing, comprising: an acquisition module, an extraction module, a construction module and an assessment module;

[0005] The acquisition module is used to acquire scene images of road safety inspection scenes and perform preprocessing to obtain image data;

[0006] The extraction module is used to extract scene features based on the image data;

[0007] The construction module is used to construct a recognition model based on the scene features;

[0008] The assessment module is used to complete the assessment simulation of the road safety inspection scenario using the recognition model.

[0009] Preferably, the acquisition module includes: an acquisition unit and a pre-processing unit;

[0010] The acquisition unit is used to acquire scene images during road inspection;

[0011] The preprocessing unit is used to preprocess the image data to obtain the image data.

[0012] Preferably, the workflow of the preprocessing unit includes:

[0013] Firstly, grayscale processing is performed on the image data to obtain a grayscale image;

[0014] Then, the grayscale image is subjected to denoising processing to obtain a denoised image;

[0015] The denoised image is resized to obtain the scene image.

[0016] Preferably, the workflow of the extraction module includes: using a bounding box detection method combined with a local binary method to perform feature extraction on the scene image to extract the scene features;

[0017] The bounding box detection method is used to extract the contour information features of the road cones, police cars and personnel in the scene image;

[0018] The local binary method is used to extract the texture features of road cones, police cars and personnel in the scene image.

[0019] Preferably, the workflow of the construction module includes: using a convolutional neural network to construct the recognition model, the recognition model includes: an input layer, four convolutional layers, a maximum pooling layer and two fully connected layers; and using ReLU and Softmax as activation functions.

[0020] The present invention also provides a road safety inspection scene assessment simulation method based on image processing, which is applied to the above system and comprises the following steps:

[0021] Collecting scene images of road safety inspection scenes and preprocessing them to obtain image data;

[0022] Extracting scene features based on the image data;

[0023] Based on the scene features, construct a recognition model;

[0024] The recognition model is used to complete the assessment simulation of the road safety inspection scenario.

[0025] Preferably, the method for performing the pretreatment comprises:

[0026] Firstly, grayscale processing is performed on the image data to obtain a grayscale image;

[0027] Then, the grayscale image is subjected to denoising processing to obtain a denoised image;

[0028] The denoised image is resized to obtain the scene image.

[0029] Preferably, the step of extracting the scene features comprises:

[0030] Using a bounding box detection method combined with a local binary method to perform feature extraction on the scene image, thereby extracting the scene features;

[0031] The bounding box detection method is used to extract the contour information features of the road cones, police cars and personnel in the scene image;

[0032] The local binary method is used to extract the texture features of road cones, police cars and personnel in the scene image.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention utilizes automation technology to reduce interference from human factors and improves the diversity and adaptability of simulation scenarios through machine learning algorithms, thereby providing law enforcement officers and staff with an assessment environment close to actual combat and reducing the manpower, material resources and time investment required for actual scenario simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] Embodiment 1

[0040] like Figure 1As shown, it is a schematic diagram of the system structure of this embodiment, including: an acquisition module, an extraction module, a construction module and an assessment module; the acquisition module is used to acquire scene images of road safety inspection scenes and perform preprocessing to obtain image data; the extraction module is used to extract scene features based on image data; the construction module is used to construct a recognition model based on scene features; the assessment module is used to use the recognition model to complete the assessment simulation of road safety inspection scenes. Among them, the acquisition module includes: an acquisition unit and a preprocessing unit; the acquisition unit is used to acquire scene images of road safety inspection scenes; the preprocessing unit is used to preprocess the scene images to obtain image data.

[0041] The following will describe in detail how the present invention solves technical problems in real life in conjunction with this embodiment.

[0042] First, the acquisition unit collects scene images during road inspections. These scene images include: image data of correctly deployed road cones, police cars and personnel, and image data of incorrectly deployed road cones, police cars and personnel. These data can come from the public data source of the traffic police department system, or they can be standard deployment images placed on site by experienced traffic police; and the image data of incorrect arrangements are annotated by professionals. These image data are then preprocessed to obtain image data. The specific steps include:

[0043] The preprocessing unit is used to convert the color image into a grayscale image to reduce processing time and data volume while retaining sufficient information for subsequent processing; then, a median filter is used to remove noise from the image, reduce noise in the image, and improve the accuracy of subsequent feature extraction and recognition; after the denoising process is completed, the denoised image is resized to meet the needs of subsequent processing to obtain image data.

[0044] After the data collection and processing is completed, the extraction module is used to extract features from the image data. The specific process includes:

[0045] First, these image data are normalized to a certain range to facilitate the processing of machine learning algorithms. Z-score normalization is used to normalize the processed images. Specifically:

[0046]

[0047] Where X represents the original data point; X min The minimum value in the data set; X max Indicates the maximum value in the data set; X normRepresents the normalized data point; thus, a normalized image is obtained. During the normalization process, if there are outliers in the data, they need to be processed before normalization to avoid excessive impact on the normalization result; at the same time, the information of the original data should be retained as much as possible during the normalization process to avoid excessive compression or loss of important features.

[0048] Meanwhile, considering the sample size of image data, this embodiment performs augmentation processing on the normalized image, wherein the steps include: random rotation, scaling, cropping, etc., to expand the data set to increase the diversity of the data and the robustness of the model.

[0049] Afterwards, the bounding box detection method is combined with the local binary method to perform feature extraction on the dataset and extract scene features.

[0050] First, the FCN model is used to segment the image data, separating the contour information features from the texture features. FCN is a deep learning model for semantic segmentation. Its main feature is the full convolution operation, which can accept input images of any size and output semantic segmentation maps of the same size. FCN replaces the fully connected layer in the traditional CNN with a fully convolutional layer, allowing the network to accept input images of different sizes and output semantic segmentation results of the corresponding size; through jump connections, FCN can effectively fuse semantic information at different levels, improving the accuracy and robustness of the segmentation results.

[0051] After that, the bounding box detection method is used to extract the contour information features of the road cones, police cars and people in the scene image. Specifically, the sliding window detection technology is used to slide a fixed-size window at different positions and scales, and each window is classified and regressed into a bounding box. After the feature extraction is completed, the principal component analysis (PCA) is used to screen out the most valuable feature subsets for individual identification (dimensionality reduction). Reducing the dimension of the feature space can improve the efficiency and accuracy of the algorithm, while reducing the computational complexity and storage requirements.

[0052] The local binary method is used to extract texture features, as follows:

[0053] For each pixel (x c ,y c ), define a circular neighborhood with a radius of R, and there are P pixels in this neighborhood, whose coordinates are (x i ,y i ), where i = 0, 1, ..., P-1.

[0054] Compare the gray value of each pixel with the central pixel I(x c ,y c) for comparison, marking pixels with higher grayscale values ​​than the center pixel as 1 and pixels with lower grayscale values ​​as 0. After the comparison is completed, a binary string consisting of P binary values ​​is obtained, which is called the local binary pattern LBP. P，R (x c ,y c ):

[0055]

[0056] Among them, LBP P，R (x c ,y c ) means (x c ,y c ) is the local binary pattern centered at I(x i ,y i ) represents the gray value of the i-th pixel in the neighborhood; I(x c ,y c ) represents the center pixel; s(·) represents the sign function, which is defined as:

[0057]

[0058] After the feature extraction is completed, a recognition module is constructed using the building module to learn the extracted features. Specifically, this network includes an input layer, four convolutional layers, a maximum pooling layer, and two fully connected layers.

[0059] The input layer is used to input the extracted feature image, with a size of W×H×3; the feature image is then input to the convolution layer for learning. The structures of these convolution layers are as follows: two convolution layers contain 64 3x3 convolution kernels, with a stride of 1 and padding of 1; one convolution layer contains 128 3x3 convolution kernels, with a stride of 1 and padding of 1; one convolution layer contains 256 3x3 convolution kernels, with a stride of 1 and padding of 1. The activation function they use is the ReLU function.

[0060] After that, these feature images will enter the maximum pooling layer to reduce the size of the feature map and reduce the amount of calculation. Specifically, the maximum pooling layer will downsample the feature map, retain the maximum value in the adjacent area, and discard other values. The maximum pooling layer uses a 2x2 pooling kernel with a step size of 2. The maximum pooling layer compresses the feature map and reduces the size of the feature map, thereby reducing the amount of calculation for subsequent convolutional layers and fully connected layers.

[0061] The fully connected layer is used to convert the feature map into a one-dimensional vector for classification or regression tasks. The structures of the two fully connected layers are as follows: one contains 1024 neurons and the activation function is ReLU, and the other contains 2 neurons and the activation function is Softmax.

[0062] After the model is built, the assessment module inputs the candidate's arrangement results into the recognition model for recognition. If the recognition result is wrong, the result is judged as unqualified, and the correctly deployed scene is automatically played for the candidate.

[0063] Embodiment 2

[0064] This embodiment also provides a road safety inspection scene assessment simulation method based on image processing, the steps comprising:

[0065] S1. Collect scene images of road safety inspection scenes and perform preprocessing to obtain image data.

[0066] First, scene images are collected during road inspections. These scene images include: image data of correctly deployed road cones, police cars and personnel, and image data of incorrectly deployed road cones, police cars and personnel. These data can come from the public data source of the traffic police department system, or they can be standard deployment images placed on site by experienced traffic police; and the image data of incorrect deployments are annotated by professionals. These image data are then preprocessed to obtain image data. The specific steps include:

[0067] The color image is converted into a grayscale image to reduce processing time and data volume while retaining sufficient information for subsequent processing; then, a median filter is used to remove noise from the image, reduce noise in the image, and improve the accuracy of subsequent feature extraction and recognition; after the denoising process is completed, the denoised image is resized to meet the needs of subsequent processing to obtain image data.

[0068] S2. Extract scene features based on image data.

[0069] First, these image data are normalized to a certain range to facilitate the processing of machine learning algorithms. Z-score normalization is used to normalize the processed images. Specifically:

[0070]

[0071] Where X represents the original data point; X min The minimum value in the data set; X max Indicates the maximum value in the data set; X norm Represents the normalized data point; thus, a normalized image is obtained. During the normalization process, if there are outliers in the data, they need to be processed before normalization to avoid excessive impact on the normalization result; at the same time, the information of the original data should be retained as much as possible during the normalization process to avoid excessive compression or loss of important features.

[0072] Meanwhile, considering the sample size of image data, this embodiment performs augmentation processing on the normalized image, wherein the steps include: random rotation, scaling, cropping, etc., to expand the data set to increase the diversity of the data and the robustness of the model.

[0073] Afterwards, the bounding box detection method is combined with the local binary method to perform feature extraction on the dataset and extract scene features.

[0074] First, the FCN model is used to segment the image data, separating the contour information features from the texture features. FCN is a deep learning model for semantic segmentation. Its main feature is the full convolution operation, which can accept input images of any size and output semantic segmentation maps of the same size. FCN replaces the fully connected layer in the traditional CNN with a fully convolutional layer, allowing the network to accept input images of different sizes and output semantic segmentation results of the corresponding size; through jump connections, FCN can effectively fuse semantic information at different levels, improving the accuracy and robustness of the segmentation results.

[0075] After that, the bounding box detection method is used to extract the contour information features of the road cones, police cars and people in the scene image. Specifically, the sliding window detection technology is used to slide a fixed-size window at different positions and scales, and each window is classified and regressed into a bounding box. After the feature extraction is completed, the principal component analysis (PCA) is used to screen out the most valuable feature subsets for individual identification (dimensionality reduction). Reducing the dimension of the feature space can improve the efficiency and accuracy of the algorithm, while reducing the computational complexity and storage requirements.

[0076] The local binary method is used to extract texture features, as follows:

[0077] For each pixel (x c ,y c ), define a circular neighborhood with a radius of R, and there are P pixels in this neighborhood, whose coordinates are (x i ,y i ), where i = 0, 1, ..., P-1.

[0078] Compare the gray value of each pixel with the central pixel I(x c ,y c ) for comparison, marking pixels with higher grayscale values ​​than the center pixel as 1 and pixels with lower grayscale values ​​as 0. After the comparison is completed, a binary string consisting of P binary values ​​is obtained, which is called the local binary pattern LBP. P，R (x c ,y c ):

[0079]

[0080] Among them, LBP P，R (x c ,y c ) means (x c ,y c ) is the local binary pattern centered at I(x i ,y i ) represents the gray value of the i-th pixel in the neighborhood; I(x c ,y c ) represents the center pixel; s(·) represents the sign function, which is defined as:

[0081]

[0082] S3. Build a recognition model based on scene features.

[0083] After feature extraction is completed, a recognition module is built to learn the extracted features. Specifically, this network consists of an input layer, four convolutional layers, a maximum pooling layer, and two fully connected layers.

[0084] The input layer is used to input the extracted feature image, with a size of W×H×3; the feature image is then input to the convolution layer for learning. The structures of these convolution layers are as follows: two convolution layers contain 64 3x3 convolution kernels, with a stride of 1 and padding of 1; one convolution layer contains 128 3x3 convolution kernels, with a stride of 1 and padding of 1; one convolution layer contains 256 3x3 convolution kernels, with a stride of 1 and padding of 1. The activation function they use is the ReLU function.

[0085] After that, these feature images will enter the maximum pooling layer to reduce the size of the feature map and reduce the amount of calculation. Specifically, the maximum pooling layer will downsample the feature map, retain the maximum value in the adjacent area, and discard other values. The maximum pooling layer uses a 2x2 pooling kernel with a step size of 2. The maximum pooling layer compresses the feature map and reduces the size of the feature map, thereby reducing the amount of calculation for subsequent convolutional layers and fully connected layers.

[0086] The fully connected layer is used to convert the feature map into a one-dimensional vector for classification or regression tasks. The structures of the two fully connected layers are as follows: one contains 1024 neurons and the activation function is ReLU, and the other contains 2 neurons and the activation function is Softmax.

[0087] S4. Use the recognition model to complete the assessment simulation of road safety inspection scenarios.

[0088] The candidate's arrangement results are input into the recognition model for recognition. When the recognition result is wrong, the result is judged as unqualified, and the correctly deployed scene is automatically played for the candidate.

[0089] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. The road safety inspection scene assessment simulation system based on image processing is characterized by: include: Acquisition module, extraction module, construction module and assessment module; The acquisition module is used to acquire scene images of road safety inspection scenes and perform preprocessing to obtain image data; The extraction module is used to extract scene features based on the image data; The construction module is used to construct a recognition model based on the scene features; The assessment module is used to complete the assessment simulation of the road safety inspection scenario using the recognition model.

2. The road safety inspection scene assessment simulation system based on image processing according to claim 1 is characterized in that: The acquisition module includes: an acquisition unit and a pre-processing unit; The acquisition unit is used to acquire scene images during road inspection; The preprocessing unit is used to preprocess the image data to obtain the image data.

3. The road safety inspection scene assessment simulation system based on image processing according to claim 2 is characterized in that: The workflow of the pre-processing unit includes: Firstly, grayscale processing is performed on the image data to obtain a grayscale image; Then, the grayscale image is subjected to denoising processing to obtain a denoised image; The denoised image is resized to obtain the scene image.

4. The road safety inspection scene assessment simulation system based on image processing according to claim 1 is characterized in that: The workflow of the extraction module includes: using a bounding box detection method combined with a local binary method to perform feature extraction on the scene image to extract the scene features; The bounding box detection method is used to extract the contour information features of the road cones, police cars and personnel in the scene image; The local binary method is used to extract the texture features of road cones, police cars and personnel in the scene image.

5. The road safety inspection scene assessment simulation system based on image processing according to claim 1 is characterized in that: The workflow of the building module includes: using a convolutional neural network to build the recognition model, the recognition model includes: an input layer, four convolutional layers, a maximum pooling layer and two fully connected layers; and using ReLU and Softmax as activation functions.

6. A road safety inspection scene assessment simulation method based on image processing, the method is applied to the system according to any one of claims 1 to 5, characterized in that the steps include: Collecting scene images of road safety inspection scenes and preprocessing them to obtain image data; Extracting scene features based on the image data; Based on the scene features, construct a recognition model; The recognition model is used to complete the assessment simulation of the road safety inspection scenario.

7. The road safety inspection scene assessment simulation method based on image processing according to claim 6 is characterized in that: The method for performing the pretreatment comprises: Firstly, grayscale processing is performed on the image data to obtain a grayscale image; Then, the grayscale image is subjected to denoising processing to obtain a denoised image; The denoised image is resized to obtain the scene image.

8. The road safety inspection scene assessment simulation method based on image processing according to claim 6 is characterized in that: The step of extracting the scene features comprises: Using a bounding box detection method combined with a local binary method to perform feature extraction on the scene image, thereby extracting the scene features; The bounding box detection method is used to extract the contour information features of the road cones, police cars and personnel in the scene image; The local binary method is used to extract the texture features of road cones, police cars and personnel in the scene image.

Images