A risk road section identification and evaluation method and system

Abstract

The application discloses a risk road section identification and evaluation method and system, and belongs to the technical field of image processing. The method specifically comprises the following steps: obtaining a target object image generated by an image sensor; preliminarily evaluating feature differences between a target object and a background in the target object image, and constructing a first coordinate transformation model, wherein the feature differences comprise texture, color and edge feature differences, and the first coordinate transformation model is constructed through preliminary evaluation of the feature differences and geometric transformation; performing coordinate deformation mapping conversion on parameters of the first coordinate transformation model to obtain a second coordinate transformation model; and performing identification and evaluation of a risk road section based on the second coordinate transformation model. Through fusion of image saliency, coordinate response and time sequence change information, the application realizes accurate positioning and classification of a target object, and is especially suitable for use in a complex background, a weak light environment or a dynamic scene.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image processing, and particularly relates to a risk road section identification and evaluation method and system. BACKGROUND

[0002] Existing target object identification methods mostly rely on static feature extraction and fixed coordinate models for identification, and when facing complex backgrounds, light changes or target object motion deformation, etc., problems such as identification accuracy decline, target positioning deviation or missed detection and misjudgment, etc. are prone to occur.

[0003] Existing technologies usually identify target objects through image processing algorithms such as edge detection, feature extraction, region segmentation, etc., but these image processing methods are often difficult to completely exclude background interference, especially for obstacle identification in dynamic environments, there are still great challenges, thereby producing large errors in risk road section identification and evaluation.

[0004] Therefore, a new method is needed to solve these problems, especially in the identification of complex road environment vehicles and roadside obstacles. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a risk road section identification and evaluation method and system.

[0006] To achieve the above purpose, the present application provides the following technical scheme:

[0007] A risk road section identification and evaluation method, comprising:

[0008] obtaining a target object image generated by an image sensor;

[0009] preliminarily evaluating the feature difference between the target object and the background in the target object image, and constructing a first coordinate transformation model, the feature difference including the difference in texture, color and edge feature, and the first coordinate transformation model being constructed through preliminary evaluation of the feature difference and geometric transformation;

[0010] deforming and mapping the first coordinate transformation model parameters to obtain a second coordinate transformation model;

[0011] based on the second coordinate transformation model, identifying and evaluating the risk road section.

[0012] Specifically, the preliminary evaluation of the feature difference between the target object and the background in the target object image and the construction of the first coordinate transformation model include:

[0013] dividing the target object image region, and extracting the features of different regions in the target object image, including texture, color and edge features;

[0014] a preliminary feature difference between the target object and the background is obtained by preliminarily evaluating a feature difference between the target object and the background;

[0015] a first coordinate transformation model is constructed by using geometric transformation according to the preliminary feature difference between the target object and the background, and the preliminary feature difference between the target object and the background is used to determine initial parameters of the coordinate transformation model.

[0016] Specifically, the target object image region is divided, and features of different regions in the target object image are extracted, including:

[0017] the target object image is preliminarily divided by using image saliency detection to generate a preliminary candidate region;

[0018] the preliminary candidate region is segmented into superpixel regions by superpixel segmentation;

[0019] texture, color and edge features of the superpixel regions are extracted by using multi-scale extraction.

[0020] Specifically, the preliminary feature difference between the target object and the background is obtained by preliminarily evaluating a feature difference between the target object and the background, including:

[0021] the superpixel regions are divided into a target object region candidate set R1 and a background region candidate set R2, and a difference score of each region in R1 and R2 with the target object image is calculated;

[0022] the difference scores of each region with the target object image are sorted, and regions corresponding to top n difference scores are reserved as the preliminary feature difference between the target object and the background.

[0023] Specifically, the preliminary coordinate transformation model is constructed by using geometric transformation according to the preliminary feature difference between the target object and the background, including:

[0024] a preliminary coordinate transformation model is constructed according to the preliminary feature difference between the target object and the background, and preliminary geometric transformation parameters are calculated;

[0025] feature points are extracted from the reserved target object region and background region, a mapping relationship between the target object region and the background region is determined, and a preliminary transformation matrix is calculated;

[0026] the preliminary transformation matrix is adjusted according to local feature differences in the reserved target object region to obtain a first transformation matrix;

[0027] the preliminary coordinate transformation model is adjusted according to the first transformation matrix to obtain a first coordinate transformation model.

[0028] Specifically, the first coordinate transformation model parameter is mapped and converted by coordinate deformation to obtain a second coordinate transformation model, comprising:

[0029] In each frame of target object image, the target object image is divided into scale levels, and a coordinate response map is constructed;

[0030] The difference gradient map is extracted by analyzing the change difference of the target object region and the background region in the displacement direction, amplitude and angle;

[0031] The coordinate response map and the difference gradient map are fused to obtain a deformation mapping candidate region;

[0032] A coordinate deformation mapping conversion function is generated for each deformation mapping candidate region, and is uniformly projected into the first coordinate transformation model to obtain the second coordinate transformation model.

[0033] Specifically, the risk road section is identified and evaluated based on the second coordinate transformation model, comprising:

[0034] According to the second coordinate transformation model, the coordinate change difference value of the target object region is calculated;

[0035] According to the coordinate change difference value of the target object region, the target object region coordinate change amplitude is calculated to generate a response intensity map;

[0036] The high-value region in the response intensity map is selected as a high-confidence object candidate area, and the image of the object is determined according to the clustering characteristics of the high-confidence object candidate area;

[0037] The object image is input into a classification neural network to obtain the category of the target object;

[0038] Based on the identified target object and the target object category, the risk road section is identified and evaluated.

[0039] Specifically, the target object comprises a vehicle and a roadside obstacle.

[0040] A risk road section identification and evaluation system is used to implement the risk road section identification and evaluation method, comprising an image acquisition module, a first model construction module, a second model construction module and a target object identification module;

[0041] The image acquisition module is used to acquire the target object image generated by the image sensor;

[0042] The first model construction module is used to preliminarily evaluate the feature difference between the target object and the background in the target object image, and construct a first coordinate transformation model;

[0043] The second model construction module is configured to perform deformation mapping on the first coordinate transformation model parameter to obtain a second coordinate transformation model.

[0044] The target object identification module is configured to identify and evaluate the risk road section based on the second coordinate transformation model.

[0045] Specifically, the first model construction module comprises a region feature extraction unit, a preliminary feature difference calculation unit and a first coordinate transformation model construction unit.

[0046] The region feature extraction unit is configured to divide the target object image region and extract the features of different regions in the target object image.

[0047] The preliminary feature difference calculation unit is configured to preliminarily evaluate the feature difference between the target object and the background to obtain the preliminary feature difference between the target object and the background.

[0048] The first coordinate transformation model construction unit is configured to construct the first coordinate transformation model by using geometric transformation according to the preliminary feature difference between the target object and the background.

[0049] Compared with the prior art, the present application has the following advantages:

[0050] The present application provides a risk road section identification and evaluation method and system, which can dynamically construct and optimize the coordinate transformation model based on the feature difference between the target object and the background, dynamically generate the response intensity map by fusing the coordinate response map, the difference gradient map and the coordinate change difference value between the image frames, effectively identify the target object in the complex background, night or shielding condition, and realize the accurate positioning and classification of the target object. Compared with the existing image processing technology, the method has higher identification accuracy in complex background, weak light environment or dynamic scene, and is especially suitable for night, shielding, multi-target and other practical applications. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A risk road section identification and evaluation method flowchart is provided for the present application.

[0052] Figure 2 A region division and feature extraction schematic diagram is provided for the present application.

[0053] Figure 3 A second coordinate transformation model schematic diagram is provided for the present application.

[0054] Figure 4 A coordinate change amplitude response intensity map is provided for the present application.

[0055] Figure 5 A risk road section identification and evaluation system architecture diagram is provided for the present application. DETAILED DESCRIPTION

[0056] The present application will be described in detail below with specific embodiments. The following examples will help those skilled in the art to further understand the present application, but in no way limit the present application. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made. These are within the scope of the present application.

[0057] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific embodiments described here are only used to explain the present application and do not limit the present application.

[0058] It should be noted that the features in the embodiments of the present application can be combined with each other without conflict, and are within the scope of protection of the present application. In addition, although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the device or the order in the flowchart. In addition, the "first", "second", "third" and the like used in the present application do not limit the data and execution order, but only distinguish the same items or similar items with basically the same function and effect.

[0059] Unless otherwise defined, all technical and scientific terms used in the specification have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not used to limit the present application. The term "and / or" used in the specification includes any and all combinations of one or more related listed items.

[0060] Example 1

[0061] Please refer to Figures 1-4 The present application provides an embodiment: a risk road section identification and evaluation method, which can be applied to the scene of dynamic detection of vehicles and obstacles in highway, can predict whether to collide in advance, realize automatic identification and evaluation of traffic risk road section, including the following specific steps:

[0062] Step S1: obtaining the target object image generated by the image sensor.

[0063] It should be noted that the image sensor is a monitoring camera with low-light imaging capability and wide dynamic range, which is suitable for image acquisition under night or complex lighting conditions. This image acquisition is not only suitable for static image acquisition, but also suitable for video stream acquisition.

[0064] Step S2: Perform a preliminary evaluation of the feature differences between the target object and the background in the target object image, and construct a first coordinate transformation model. The feature differences include differences in texture, color, and edge features. The first coordinate transformation model is constructed through preliminary evaluation of feature differences and geometric transformation.

[0065] It should be noted that the target objects include, but are not limited to, vehicles, roadside obstacles, pedestrians and other obstacles. In the highway environment, this application is particularly applicable to the identification and detection of vehicles and roadside obstacles, and to risk identification based on this.

[0066] The specific steps of step S2 are as follows:

[0067] Step S201: Divide the target object image region and extract the features of different regions in the target object image, including: texture, color and edge features.

[0068] like Figure 2 As shown, a highway scene against a complex nighttime background is provided. Step S201 is used to perform region segmentation and feature extraction on the target object image. The specific steps are as follows:

[0069] Step S2011: Use image saliency detection to perform preliminary segmentation of the target object image and generate preliminary candidate regions.

[0070] First, an image of the target object is acquired, such as... Figure 2 As shown in the upper left corner, the image contains multiple vehicle targets and a complex nighttime background, including streetlights, trees, and headlight spots.

[0071] On the other hand, based on image saliency detection, saliency analysis is performed on the target object image to extract the region of interest in the image. Through this saliency analysis, multiple preliminary candidate regions are generated, such as... Figure 2 As shown in the upper right corner, these areas are considered to contain potential target objects or their edge contours, or are background areas. The preliminary candidate areas have strong information density and contrast.

[0072] Step S2012: Perform superpixel segmentation on the preliminary candidate region to divide it into superpixel regions.

[0073] Subsequently, each preliminary candidate region is further refined. In this embodiment, superpixel segmentation is used to generate multiple image patches with high structural consistency. Each superpixel region has high color consistency and edge continuity.

[0074] like Figure 2 As shown in the lower right region, the initial candidate region can be subdivided into several irregular, locally consistent image regions. Figure 2The more detailed segmentation is not shown, and the super-pixel division is more accurate than the traditional rectangular division method, and is suitable for extraction of the boundary of a complex obstacle shape.

[0075] Step S2013: Texture, color and edge features of the super-pixel region are extracted using multi-scale.

[0076] In this step, multi-scale feature extraction is performed on all super-pixel regions. It should be noted that, using the prior art method, local binary pattern and gray level co-occurrence matrix method are used to extract texture features in the super-pixel region, including texture distribution, texture contrast and texture directionality; the pixels in the super-pixel region are converted to HSV space, the color histogram is calculated and normalized, and the color features are extracted; the Canny operator is used to detect the boundary information, and the Sobel gradient is combined to calculate the edge density of the region, and the edge features of the super-pixel region are extracted to model the boundary between the obstacle contour and the background.

[0077] Through the step S201 described in this embodiment, the distinguishability of the background region and the target object region can be effectively improved under the premise of ensuring the consistency of the region.

[0078] Step S202: Preliminary evaluation of the feature difference between the target object and the background is performed to obtain the preliminary feature difference between the target object and the background.

[0079] The specific steps of step S202 are as follows:

[0080] Step S2021: The super-pixel region is divided into a target object region candidate set R1 and a background region candidate set R2, and the difference score of each region in R1 and R2 with the target object image is calculated.

[0081] In this embodiment, after the super-pixel region feature extraction is completed, all super-pixel regions are divided into two sets according to their saliency response, color distribution, position deviation and other indicators: the target object region candidate set R1, which includes super-pixel regions similar in color and texture to the obstacle, and edge feature dense super-pixel regions; the background region candidate set R2, which includes image edge regions, low saliency response regions, or regions with significantly different texture distribution and obstacle features.

[0082] The difference score of each region in R1 and R2 with the target object image region preliminarily positioned in the entire image is calculated, and the difference score is based on texture difference, color difference and edge structure difference. Each difference score comprehensively reflects the overall difference between the current region and the target object in texture, color and edge.

[0083] Step S2022: Sort the difference scores of each region with the target object image, and retain the top n regions corresponding to the difference scores as the preliminary feature difference between the target object and the background.

[0084] In this embodiment, n = 5 is set as an example, and the selected five regions include the car light, the car tail contour, the high-contrast edge, and part of the background interference region. The region with significant feature difference can be selected from the multiple candidate regions, and the target object region and the background region can be further distinguished.

[0085] Step S203: According to the preliminary feature difference between the target object and the background, determine the initial parameters of the coordinate transformation model, and construct a first coordinate transformation model by using geometric transformation.

[0086] The specific steps of step S203 are as follows:

[0087] Step S2031: According to the preliminary feature difference between the target object and the background, construct a preliminary coordinate transformation model, and calculate the preliminary geometric transformation parameters;

[0088] In this embodiment, affine transformation or perspective transformation is introduced as the form framework of the preliminary coordinate transformation model. The initial control point pairs required by the transformation model are derived by using the preliminary feature difference between the target object and the background, such as the center point of the car light, the tail edge line, and the background street lamp line, and a set of geometric transformation parameters are calculated accordingly.

[0089] Step S2032: Extract feature points from the retained target object region and background region, determine the mapping relationship between the target object region and the background region, and calculate the preliminary transformation matrix;

[0090] In this step, feature points are extracted from the candidate target object region and the background region to establish the spatial mapping relationship. The scale-invariant feature transform (SIFT) algorithm is used to detect key points, and the point pair matching is performed by combining the image matching algorithm. It should be noted that the selection of the feature point pair should give priority to ensuring that the point pair has obvious difference, such as the tail light and the road lamp pole. By using the feature point pair, the affine or perspective relationship between the point sets is constructed, and the preliminary transformation matrix is calculated to realize the alignment of the target object region and the background region in the geometric space.

[0091] Step S2033: According to the local feature difference in the retained target object region, adjust the preliminary transformation matrix to obtain a first transformation matrix;

[0092] In the embodiment, the error caused by background disturbance or target deformation exists in the preliminary calculated transformation matrix, a local optimization strategy is introduced, the feature difference correction of a reserved target object area is performed in a fine granularity, specifically, the transformation error of the feature points in each area is calculated, a local affine disturbance factor is introduced, the parameters in the preliminary transformation matrix are fine-tuned, the transformation matrix is iteratively optimized with the minimum coordinate residual in the area as the target, and finally the first transformation matrix is obtained.

[0093] Step S2034: adjusting the preliminary coordinate transformation model according to the first transformation matrix to obtain a first coordinate transformation model.

[0094] In the embodiment, the first transformation matrix is set as T, the first coordinate transformation model M(x, y) is constructed, and then The three-dimensional coordinates of the target object are obtained by extending the two-dimensional coordinates (x, y), and the purpose is to include the translation transformation in the matrix operation. If only two-dimensional coordinates are used, only linear transformations such as rotation, scaling, and symmetry can be expressed, but translation cannot be expressed. It should be noted that the content of the target object image is dynamically updated.

[0095] Step S3: deforming and mapping the first coordinate transformation model parameters to obtain a second coordinate transformation model.

[0096] As shown in Figure 3 For the slight deformation, posture change or background interference of the target object in the target object image in the time sequence image, the second coordinate transformation model is further constructed by modeling the difference between the response change inside the target object region and the background interference on the basis of the first coordinate transformation model, so as to enhance the adaptability of target positioning in a complex environment.

[0097] The specific steps of step S3 are as follows:

[0098] Step S301: in each frame of target object image, the target object image is divided into scale levels, and a coordinate response map is constructed.

[0099] On the one hand, in order to adapt to the deformation of vehicles of different sizes in the image, in the embodiment, the image pyramid strategy is adopted to divide each frame of image into multiple scale levels, for example, the original size, 0.75 times, 0.5 times, etc., and the coordinate response modeling is performed separately at each scale.

[0100] On the other hand, in each scale level, according to the position change of the target object region and the background region after the coordinate transformation, a coordinate response map is constructed. Each pixel value in the response map represents the response intensity of the position under the first coordinate transformation model, for example, the coordinate offset, the residual size, the larger the value is, the more sensitive the region is to the coordinate transformation.

[0101] Exemplarily, as in the night target object image, the car light region presents consistent and high coordinate response values at multiple scale levels due to high brightness and clear edges, while the background tree shadow presents low or no response.

[0102] Step S302: Analyze the difference in displacement direction, amplitude, and angle between the target object region and the background region, and extract a difference gradient map.

[0103] Specifically, the displacement direction, amplitude, and angle of the target object region and the background region in adjacent image frames are analyzed to construct a difference gradient map, including: displacement direction, indicating the transformation direction vector; displacement amplitude, indicating the intensity difference of the coordinate response of each region; and angle gradient, reflecting the angle twist trend after transformation.

[0104] In this way, the sensitivity of the target object region to deformation can be extracted from the structure, and random changes caused by background noise can be distinguished.

[0105] Step S303: Fuse the coordinate response map and the difference gradient map to obtain a deformation mapping candidate region;

[0106] The fusion method uses weighted superposition or pixel-level multiplication enhancement to enhance the response intensity of the deformation region, and finally obtains the deformation mapping candidate region.

[0107] In this embodiment, the regions where the response value and the gradient difference both exceed the set threshold are exemplarily selected as the candidate regions, which are mostly concentrated in the car tail edge, light region, and contour region.

[0108] Step S304: Generate a coordinate deformation mapping conversion function for each deformation mapping candidate region, and project it into the first coordinate transformation model to obtain a second coordinate transformation model.

[0109] Specifically, for each deformation mapping candidate region, a local coordinate transformation function is constructed based on the local coordinate response of the feature points in the region. The function can be a local affine, local perspective, or non-linear function based on a spline f i (x,y), and all local transformation functions f i are projected into the first coordinate transformation model M for fusion, and the specific formula is: M1(x,y) represents the second coordinate transformation model, ω i represents the weight of the i-th local coordinate transformation function, which is dynamically assigned according to the region response intensity and recognition confidence, and I represents the number of local coordinate transformation functions.

[0110] Finally, the second coordinate transformation model M1 is obtained, as shown in Figure 3As shown, the coordinate transformation model not only contains the original global coordinate mapping relationship, but also integrates the response characteristics of local micro-deformation, and can more accurately reflect the real position and posture of the target object in a complex scene.

[0111] Step S4: Based on the second coordinate transformation model, the risk road section is identified and evaluated.

[0112] The specific steps of step S4 are:

[0113] Step S401: According to the second coordinate transformation model, the coordinate change difference value of the target object region is calculated;

[0114] Specifically, in the time sequence, each frame of target object image uses the second coordinate transformation model, and the target object region is mapped to the standard coordinate system. In the t-th frame and the t+1-th frame, the key point coordinates in the target object region are tracked and compared to obtain the coordinate change difference value between frames.

[0115] Step S402: According to the coordinate change difference value of the target object region, the coordinate change amplitude of the target object region is calculated, and a response intensity map is generated;

[0116] The coordinate change difference value obtained in the above step is subjected to a modulus operation to obtain the coordinate change amplitude of the target object region, such as Figure 4 As shown, the white bright closed loop area in the lower left represents the vehicle tail area, which presents continuous displacement and shape change in multiple images, has a large response amplitude, the small bright block in the middle is a distant vehicle, and the point-like bright spot in the upper right is a background street lamp, which has a small response amplitude and is sparsely distributed as a point, because the street lamps are generally evenly distributed, so they are relatively static, resulting in a small response amplitude.

[0117] Step S403: Select a high-value region in the response intensity map as a high-confidence object candidate area, and determine the object image according to the clustering characteristics of the high-confidence object candidate area;

[0118] In an embodiment, a response intensity threshold is set, a high-response region is extracted, and a clustering analysis method is used to identify a concentrated and closed high-response block, such as Figure 4 As shown, the largest bright area block is selected as the candidate area of the main object, for example, a vehicle, while the point-like area is excluded or processed with low weight, for example, evenly distributed street lamps or trees, etc. It should be noted that, Figure 4 What is provided is a relatively uniform and obvious example.

[0119] It should be noted that the target object image is an image containing a target object, a background, and an obstacle, etc., and the object image only contains a target object.

[0120] Step S404: input the object image into the classification neural network to obtain the category of the target object;

[0121] In this embodiment, a lightweight convolutional neural network can be used. After the lightweight convolutional neural network is trained, the object image is input to obtain the category of the target object.

[0122] Step S405: identify and evaluate the risk road section based on the identified target object and the category of the target object.

[0123] It should be noted that, in this embodiment, the identified category of the target object can be used not only for the basic vehicle identification task, but also can be further combined with traffic scene information to realize automatic identification and evaluation of the traffic risk road section.

[0124] Specifically, in the continuous video frames, coordinate tracking and response intensity monitoring are performed on the identified target object, and an information sequence is constructed, including: a target motion trajectory, recording the position change of the target object in each frame under the second coordinate transformation model; a response intensity evolution graph, recording the response amplitude change of the target object region in each frame; and a dynamic aggregation degree index, calculating the regional dynamic aggregation characteristics according to the motion trend, distance convergence degree and direction consistency of multiple vehicle targets.

[0125] On the other hand, high-risk behavior features are extracted; when any one or more of the following features appears in the same space region, the region is marked as a high-risk candidate road section, and the features include: the vehicle region always shows a strong coordinate response in continuous frames, such as frequent hard braking, deflection and pause; the target trajectory is distorted or crossed, and the trajectories of multiple vehicles abnormally overlap, intersect or suddenly approach in space; the region response diffusion increases, and the response graph shows a trend of diffusion from a single point to the surrounding, suspected accident or sudden aggregation; the inter-frame response mutates, and the response distribution jumps significantly between frames, which is determined to be caused by sudden behavior, such as lane changing and collision; and it is determined that the high-risk candidate road section has traffic risk.

[0126] For example, in the image sequence, if the system continuously detects that: the rear region response value is high; the rear vehicle frequently approaches; multiple targets suddenly slow down and gather in the intersection region; the position corresponding to the image is marked as a traffic risk road section region.

[0127] Embodiment 2

[0128] Please refer to Figure 2 Another embodiment provided by the present application is a risk road section identification and evaluation system, comprising: an image acquisition module, a first model construction module, a second model construction module and a target object identification module.

[0129] The image acquisition module is configured to acquire a target object image generated by an image sensor.

[0130] The first model construction module is configured to preliminarily evaluate the feature difference between the target object and the background in the target object image, and construct a first coordinate transformation model.

[0131] The second model construction module is configured to perform deformation mapping on the parameters of the first coordinate transformation model to obtain a second coordinate transformation model.

[0132] The target object identification module is configured to identify and evaluate the risk road section based on the second coordinate transformation model.

[0133] The first model construction module comprises a region feature extraction unit, a preliminary feature difference calculation unit, and a first coordinate transformation model construction unit.

[0134] The region feature extraction unit is configured to divide the target object image region and extract the features of different regions in the target object image.

[0135] The preliminary feature difference calculation unit is configured to preliminarily evaluate the feature difference between the target object and the background to obtain the preliminary feature difference between the target object and the background.

[0136] The first coordinate transformation model construction unit is configured to construct the first coordinate transformation model by using geometric transformation according to the preliminary feature difference between the target object and the background.

[0137] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive repetition.

[0138] The specific embodiments described above further explain the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A risk road section identification and evaluation method, characterized in that, The method comprises the following steps: obtaining an image of a target object generated by an image sensor; preliminarily evaluating feature differences between the target object and the background in the image of the target object, and constructing a first coordinate transformation model, wherein the feature differences include texture, color and edge feature differences, and the first coordinate transformation model is constructed by preliminary evaluation of the feature differences and geometric transformation; performing coordinate deformation mapping conversion on parameters of the first coordinate transformation model to obtain a second coordinate transformation model, wherein the coordinate deformation mapping conversion obtains a deformation mapping candidate region by fusing a coordinate response map and a difference gradient map, and generates a coordinate deformation mapping conversion function by using the deformation mapping candidate region to perform conversion; the coordinate response map is obtained by dividing the image of the target object into scale levels, and in each scale level, a position change after coordinate transformation of the target object region and the background region is obtained; the difference gradient map is obtained by analyzing displacement direction, amplitude and angle change of the target object region and the background region in adjacent image frames; based on the second coordinate transformation model, identifying and evaluating a risk road section; preliminarily evaluating feature differences between the target object and the background in the image of the target object, and constructing a first coordinate transformation model, comprising: dividing the target object image region, and extracting features of different regions in the target object image, including texture, color and edge features; preliminarily evaluating feature differences between the target object and the background to obtain preliminary feature differences between the target object and the background; constructing the first coordinate transformation model by using geometric transformation according to the preliminary feature differences between the target object and the background, wherein the preliminary feature differences between the target object and the background are used to determine initial parameters of the coordinate transformation model; the division of the target object image region and the extraction of features of different regions in the target object image comprise: preliminarily dividing the target object image using image saliency detection to generate a preliminary candidate region; performing superpixel segmentation on the preliminary candidate region to divide it into superpixel regions; extracting texture, color and edge features of the superpixel regions using multi-scale; the preliminary evaluation of feature differences between the target object and the background to obtain preliminary feature differences between the target object and the background comprises: dividing the superpixel regions into a target object region candidate set R1 and a background region candidate set R2, and calculating a difference score of each region in R1 and R2 with a preliminarily positioned target object image region in the target object image, wherein the difference score is scored based on texture difference, color difference and edge structure difference; sorting the difference scores of each region with the target object image, and retaining regions corresponding to top n difference scores as preliminary feature differences between the target object and the background; the construction of the first coordinate transformation model by using geometric transformation according to the preliminary feature differences between the target object and the background comprises: constructing a preliminary coordinate transformation model according to the preliminary feature differences between the target object and the background, and calculating preliminary geometric transformation parameters; extracting feature points from the retained target object region and background region, determining a mapping relationship between the target object region and the background region, and calculating a preliminary transformation matrix; Adjust the preliminary transformation matrix according to the local feature difference in the reserved target object region, to obtain a first transformation matrix; Adjust the preliminary coordinate transformation model according to the first transformation matrix, to obtain a first coordinate transformation model; Perform coordinate deformation mapping conversion on the first coordinate transformation model parameters, to obtain a second coordinate transformation model, including: In each frame of target object image, divide the target object image into scale levels, and construct a coordinate response graph; Analyze the change difference of the target object region and the background region in the displacement direction, amplitude and angle, and extract a difference gradient graph; Fuse the coordinate response graph and the difference gradient graph, to obtain a deformation mapping candidate region; Generate a coordinate deformation mapping conversion function for each deformation mapping candidate region, and project it into the first coordinate transformation model, to obtain the second coordinate transformation model; The risk road section is identified and evaluated based on the second coordinate transformation model, including: According to the second coordinate transformation model, the coordinate change difference value of the target object region is calculated; According to the coordinate change difference value of the target object region, the target object region coordinate change amplitude is calculated, and a response intensity graph is generated; Select a high-value region in the response intensity graph as a high-confidence object candidate area, and determine the object image according to the aggregation characteristics of the high-confidence object candidate area; Input the object image into a classification neural network to obtain the category of the target object; Based on the identified target object and the category of the target object, the risk road section is identified and evaluated.

2. The method of risk road segment identification and assessment of claim 1, wherein, The target object includes a vehicle and a roadside obstacle.

3. A risk road section identification and evaluation system for implementing the risk road section identification and evaluation method of any one of claims 1-2, characterized in that, Including: An image acquisition module, a first model construction module, a second model construction module and a target object identification module; The image acquisition module is configured to acquire a target object image generated by an image sensor; The first model construction module is configured to preliminarily evaluate the feature difference between the target object and the background in the target object image, and construct a first coordinate transformation model; The second model construction module is configured to perform deformation mapping on the first coordinate transformation model parameters, to obtain a second coordinate transformation model; The target object identification module is configured to identify and evaluate the risk road section based on the second coordinate transformation model.

4. The risk road segment identification and assessment system of claim 3, wherein, The first model construction module includes a region feature extraction unit, a preliminary feature difference calculation unit and a first coordinate transformation model construction unit; The region feature extraction unit is configured to divide the target object image region and extract the features of different regions in the target object image; The preliminary feature difference calculation unit is configured to preliminarily evaluate the feature difference between the target object and the background, to obtain the preliminary feature difference between the target object and the background; The first coordinate transformation model construction unit is configured to construct a first coordinate transformation model using geometric transformation according to the preliminary feature difference between the target object and the background.

Images