Target detection system facing snowy weather and driving conflict risk assessment method

Abstract

The invention belongs to the field of road vehicle control systems, and particularly relates to a snowy weather oriented target detection system and a driving conflict risk assessment method, the target detection system takes YOLOv11n as a basic framework, and an improved detail enhancement module is embedded between a first Conv module and a second Conv module of a YOLOv11n model backbone network; a convolution module in a C3K2 module of the neck network of the YOLOv11n model is replaced by a gating bottleneck convolution module; the data processing flow of the improved detail enhancement module can be summarized as follows: parallel differential extraction is firstly carried out, self-adaptive output adjustment is carried out, equivalence re-parameterization is then carried out, and end residual connection information is enhanced; the gating bottleneck convolution module captures multi-scale frequency information of an image by using a wavelet transform convolution network, improves the perception ability of the model to the whole environment, and adaptively adjusts information flow, strengthens important features and inhibits invalid redundancy according to input content by using a gating mechanism. The technical problems of low vehicle target detection precision and high small target omission ratio in snowy weather are solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of road vehicle control systems, and relates to the calculation or prediction of driving parameters, in particular to a target detection system for blizzard weather and a driving conflict risk assessment method based on the system. BACKGROUND

[0002] With the rapid development of intelligent transportation systems (ITS) and automatic driving technology, environmental perception and driving safety warning capability have become the key link to ensure the reliable operation of automatic driving systems. Among them, the accurate detection of vehicle targets and the real-time risk assessment of traffic conflicts as core functional modules directly affect the decision planning and active safety performance of vehicles. At present, the mainstream intelligent vehicles generally use visual sensors as the main means of environmental information acquisition. However, in actual road running scenarios, complex weather conditions such as blizzards will degrade the image quality obtained by visual sensors, mainly manifested as decreased contrast, blurred target edges, and increased noise level, which will bring serious challenges to subsequent environmental perception and real-time driving warning.

[0003] Current mainstream target detection algorithms, such as the YOLO series and Faster R-CNN, have shown high detection accuracy and inference efficiency in traffic scenes with good lighting and high visibility. However, most of these methods are based on clear image assumptions for model design, and their feature extraction mechanisms have weak adaptability to low-contrast, high-noise blizzard weather images, with problems such as severe loss of detail information, high small target miss detection rate, etc. Although some studies have attempted to improve model robustness through image enhancement preprocessing or improved network structure, most methods still lack the ability to model the physical degradation process under blizzard weather, making it difficult to achieve stable and reliable vehicle perception in complex environments. At the same time, for the detected targets, monocular vision-based depth estimation still faces challenges in complex traffic scenes. Traditional convolutional neural networks have limited receptive fields, making it difficult to effectively model long-range spatial dependencies, resulting in problems such as discontinuous depth estimation at object boundaries and blurred contours. Although some mainstream methods introduce multi-scale feature fusion or global attention mechanisms to enhance context modeling capabilities, they often come with high computational complexity, making it difficult to meet the strict real-time requirements in vehicle-mounted environments. And existing methods generally lack explicit modeling of detail residual information in the decoding process, resulting in the inability to fully reuse edge and texture information in shallow high-resolution features, which can easily cause local depth distortion, affecting the accuracy and reliability of distance estimation, and making it difficult to achieve accurate environmental perception in complex environments.

[0004] In addition, the existing traffic conflict analysis methods mostly rely on the trajectory data or geographic information system (GIS) for offline statistics and comprehensive evaluation in a post-mortem manner, and the data sources are mostly historical traffic flow records or fixed monitoring equipment, lacking deep integration with the vehicle-mounted real-time sensing system. Such methods usually cannot meet the low-delay and high-response pre-warning requirements of automatic driving or advanced driver assistance system (ADAS), and are difficult to provide instant risk prompt and decision support for drivers or automatic driving controllers in a dynamic traffic environment. SUMMARY

[0005] In view of the shortcomings and deficiencies of the prior art, the first object of the present application is to provide a target detection system for blizzard weather, which takes YOLOv11n as the basic framework, introduces an improved detail enhancement module (DEA-Net) and a gating bottleneck convolution (GB_Block) module, enhances the feature extraction capability and small target perception performance in low contrast and high noise environment, and solves the technical problems of low vehicle target detection accuracy and high small target missing detection rate in blizzard weather.

[0006] To achieve the above object, the technical scheme adopted by the present application is as follows: A target detection system for blizzard weather is obtained by training a YOLO-Blizzard network model, wherein the YOLO-Blizzard network model takes a YOLOv11n model as a basic network model, and embeds an improved detail enhancement module between a first Conv module and a second Conv module of a backbone network of the YOLOv11n model; and replaces a convolution module in a C3K2 module of a neck network of the YOLOv11n model with a gating bottleneck convolution module. The improved detail enhancement module includes one standard convolution and four difference convolutions, first extracts features from input data through five parallel independent branches, then integrates the extracted features by weighting; then adds the filter kernels of the five independent branches to obtain an equivalent kernel, and the equivalent kernel and the features obtained by weighting integration are convolved again; finally, a residual connection is introduced and processed through a normalization layer. The gating bottleneck convolution module first captures the multi-scale frequency information of the image by using a wavelet transform convolution network, and the output features of the wavelet transform convolution network are processed by two cross-parallel data processing modules, the first data processing module is composed of a first CGT convolution module, a second CGT convolution module arranged in sequence, the second data processing module is composed of a third CGT convolution module, a fourth CGT convolution module arranged in sequence, the output of the second CGT convolution module and the third CGT convolution module is subjected to Hadamard product operation to obtain the input of the fourth CGT convolution module, then the output features of the fourth CGT convolution module and the output features of the wavelet transform convolution network are subjected to element-by-element addition operation to complete feature fusion; finally, the fused features are subjected to nonlinear transformation, and are connected with the original input features through residual connection to obtain the final output , the expression is: ; Wherein, K is the output feature of the fourth CGT convolution module, is the output feature of the wavelet transform convolution network, represent the identity residual without any transformation, i.e. the original input feature.

[0007] As a preferred embodiment of the application, the four difference convolutions in the improved detail enhancement module are central difference convolution, angle difference convolution, horizontal difference convolution and vertical difference convolution; when the extracted features are weighted and integrated, first, the Softmax normalization is performed on each branch feature in the channel dimension to obtain branch-specific weights, and then the corresponding features are weighted with the weights and summed along the branch.

[0008] As a preferred embodiment of the application, the first CGT convolution module, the second CGT convolution module, the third CGT convolution module and the fourth CGT convolution module have the same network structure, and each includes a convolution layer, a normalization layer and a TeLU activation function arranged in sequence.

[0009] The second object of the application is to provide a driving conflict risk assessment method for severe weather, which uses the above-mentioned target detection system for severe weather to accurately identify vehicles, and based on high-precision target detection, a self-designed lightweight depth estimation convolution network is used to accurately estimate the distance of surrounding vehicles, and the relative speed is calculated combined with tracking information; finally, based on the obtained distance and speed features, a dynamic conflict risk assessment model is constructed by using a clustering algorithm, and vehicle control suggestions are given to vehicles under high risk, providing real-time risk prompts and decision support for drivers or automatic driving controllers.

[0010] To achieve the above object, the application adopts the following technical scheme: The driving conflict risk assessment method for severe weather comprises the following steps: Step B1: Real-time receiving data collected by the vehicle-mounted camera; Step B2: Using the target detection system for blizzard weather mentioned above to detect targets in each frame of image; at the same time, calling the trained deep estimation convolution network to output the deep estimation graph corresponding to each frame of input image; wherein the deep estimation convolution network is a monocular depth estimation network, which adopts an encoder-bottleneck-decoder structure, embeds multi-scale cross attention in the bottleneck stage for multi-scale context aggregation, and realizes spatial resolution recovery through transposed convolution and residual fusion in the decoding stage; Step B3: For each target vehicle detected around the vehicle, based on the position of the target vehicle in the image and the corresponding depth, the real-time distance of each target vehicle in the current frame is calculated and output, and the speed is calculated in combination with the tracking information; Step B4: Using OBD-II interface to access the speed data of the vehicle in real time; inputting the calculated distance and speed data of the surrounding vehicles and the speed data of the vehicle into the trained clustering risk assessment model to perform clustering analysis on the current driving state and judge the conflict risk category of the vehicle in real time.

[0011] As a preferred embodiment of the present application, the method for calculating the speed of the target vehicle in combination with the tracking information in step B3 is: tracking the same target vehicle in consecutive video frames to obtain its spatial coordinates at consecutive time points; based on the spatial displacement and time interval of the target vehicle between consecutive frames, the relative motion speed of the target vehicle relative to the vehicle is calculated.

[0012] As a preferred embodiment of the present application, the encoder in the deep estimation convolution network is composed of 4 layers of convolution-pooling units, which gradually extract abstract features and reduce spatial resolution; wherein the first layer of convolution-pooling unit is: The specific operation of the first layer of convolution-pooling unit is: In the formula, is a 3*3 convolution, and the number of output channels is 64, 128, 256, and 512 in turn; is batch normalization; is a self-defined activation function; The spatial size is reduced by half.

[0013] As a preferred embodiment of the present application, the decoder in the deep estimation convolution network is composed of 4 layers of transposed convolution-residual units, each of which doubles the spatial resolution, and through the bilinear interpolation aligned skip connection, the high-resolution details reserved in the encoding stage are injected, and the kth layer of transposed convolution-residual operation is: ; In the formula, is a 4*4 transposed convolution with a step of 2; This represents upsampling, using bilinear interpolation to recover the size. The features obtained during the encoding stage This represents the normalization of each layer.

[0014] As a preferred embodiment of the present invention, the multi-scale cross-attention embedded in the bottleneck stage adopts a dual-path parallel structure, with the input features fed into two independent processing branches: First, the upper branch, after standardization, extracts the lateral spatial dependencies of different scales by applying 1×7, 1×11, and 1×21 deep convolutional kernels in parallel; the lower branch, after standardization, captures the vertical spatial dependencies of multiple scales by applying 7×1, 11×1, and 21×1 deep convolutional kernels in parallel. Second, within each branch, the output undergoes cross-channel information interaction and integration via 1×1 convolution, and is then processed through linear... The transformation generates query Q, key K, and value V matrices. Then, the Q matrix generated by the upper branch (Fx) and the Q matrix generated by the lower branch (Fy) are interchanged. The interchanged Q matrix is ​​multiplied by the corresponding K matrix and then by the V matrix to generate attention feature maps for the upper and lower branches respectively. Next, the two branch attention feature maps are rearranged in shape and channel blending is performed through 1×1 convolution. Finally, the output Fup of the upper branch and the output Fdown of the lower branch are fused to enhance global spatial context information, thereby improving the perception accuracy of object boundaries and structural details.

[0015] As a preferred embodiment of the present invention, when training the depth estimation convolutional network, the total loss function consists of data loss and gradient loss, and its expression is: ; in, It is the total loss function that the network needs to minimize during the training process; Depth map predicted by the network; : A true depth map; Data loss measures the overall numerical difference between the predicted depth value and the actual depth value. Gradient loss measures the difference between the predicted depth map and the true depth map at edges; α and β are two balancing factors used to adjust... and Weight in total loss.

[0016] As a further preferred embodiment of the present invention, the expression for data loss is: ; in, It is the first The difference between the predicted depth and the true depth of each effective pixel in logarithmic space. It is all valid pixels The sum of squares, It is all the sum of squares of the differences between the predicted depth map and the ground truth map, is a set of valid pixels in the depth map, is the total number of valid pixels; is a balance factor; The expression of gradient loss is: wherein, and are the predicted depth map are the horizontal and vertical gradient values of the predicted depth map at the i-th pixel; represents the ground truth map; is the L1 norm difference between the predicted depth map and the interpolated ground truth map in the horizontal direction gradient, is the L1 norm difference in the vertical direction gradient.

[0017] Compared with the prior art, the present application has the following advantages and beneficial effects: (1) The present application optimizes the structure of the DEA-Net module, introduces channel splicing and convolution compression mechanism to enhance the fusion expression of multi-path features, and realizes adaptive weighted fusion of branch features by combining with the gate weight suture structure, which improves the feature extraction capability of the model for blurred and low-contrast images in snow weather, thereby enhancing the detection robustness and precision of the YOLO model.

[0018] (2) The present application proposes a GB_Block module, which uses wavelet transform convolution to improve the perception ability of the model for the overall environment, and uses a gating mechanism to adaptively adjust the information flow according to the input content, strengthen important features, suppress invalid redundancy, and improve the perception ability of the target form, solving the technical problems of low vehicle target detection precision and high small target missing rate in snow weather.

[0019] (3) The present application proposes a monocular depth estimation network named DepthEstimationCNN, which adopts an encoder-bottleneck-decoder structure and integrates residual connection, multi-scale cross-attention mechanism and step-by-step upsampling recovery strategy, enhancing the modeling ability of the model for spatial context information and the reuse ability of detailed features. Compared with traditional convolutional neural networks and mainstream depth estimation frameworks, the proposed model improves the precision of depth estimation on the basis of lightweight design with relatively low parameter quantity (about 4M), and especially shows strong performance in object boundary preservation and context understanding in complex scenes.

[0020] ​​(4) The application provides a YOLO-Blizzard target detection network model, aiming to improve the recognition performance of YOLO model on vehicles, pedestrians and other targets in blizzard weather environment. The model effectively suppresses snow noise interference and enhances image quality by introducing an improved DEA-Net module. At the same time, a new GB_block structure is proposed to further strengthen the feature extraction and fusion capability. Experimental results show that the YOLO-Blizzard model significantly improves the target detection accuracy in blizzard scenes, with an increase of 6.8% in F1-Score index, achieving more accurate and robust vehicle recognition effect.

[0021] (5) The application combines real-time detection, ranging, speed measurement and clustering analysis to build a vehicle-mounted closed-loop system, realizes dynamic evaluation of real-time traffic conflict risk, and provides different suggestions and control strategies for different risk categories, thereby providing a new solution for automatic driving warning system. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The YOLO-Blizzard network model provided by the application is shown in the figure; Figure 2 The improved detail enhancement module (DEA-Net module) provided by the application is shown in the figure; Figure 3 The GB_block module provided by the application is shown in the figure; Figure 4 The driving conflict risk assessment method for blizzard weather provided by the application is shown in the figure; Figure 5 The depth estimation convolutional network (Depth Estimation CNN) provided by the application is shown in the figure; Figure 6 The MSCAAttention network provided by the application is shown in the figure. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solutions of the application and their advantages, the application will be described in detail below with reference to the accompanying drawings, but not used to limit the protection scope of the application.

[0024] Example 1:

[0025] As Figure 1As shown, the embodiment provides a target detection system for blizzard weather, which is obtained by training a YOLO-Blizzard network model, wherein the YOLO-Blizzard network model is based on a YOLOv11n model as a basic network model, and an improved detail enhancement module (DEA-Net module) is embedded between a first Conv module and a second Conv module of a backbone network of the YOLOv11n model; and a convolution module in a C3K2 module of a neck network of the YOLOv11n model is replaced with a gated bottleneck convolution module (GB_block module); The improved detail enhancement module (DEA-Net module) is composed of a standard convolution (Vanilla Conv, VC) and four kinds of differential convolutions, and the main steps are as follows: first, the input is feature-extracted through five convolution modules, and the extracted features are weighted using Softmax to improve the representation ability of the features; second, the five convolutions are added to obtain an equivalent convolution (equivalent convolution) using the convolution additivity, and the equivalent convolution and the features obtained by Softmax are convolved again to realize fine enhancement of the key features; and finally, a residual connection is introduced to prevent information loss; The gated bottleneck convolution module (GB_block module) uses wavelet transform convolution to capture multi-scale frequency information of the image, enhances the perception ability of the model to the overall environment and the fine structure of the target (including edges, contours, etc.), and at the same time, uses a gating mechanism to adaptively adjust the input features, strengthens important features and suppresses redundant noise, thereby improving the small target perception ability and robustness of the model in a complex background.

[0026] As shown, in the embodiment, the data processing flow of the improved detail enhancement module (DEA-Net) can be summarized as "first parallel differential extraction, second adaptive adjustment output, third equivalent reparameterization, and fourth residual connection information enhancement", and the specific steps are as follows: Figure 2 As shown, in the embodiment, the data processing flow of the improved detail enhancement module (DEA-Net) can be summarized as "first parallel differential extraction, second adaptive adjustment output, third equivalent reparameterization, and fourth residual connection information enhancement", and the specific steps are as follows: A group of input images is given : ; Wherein, B represents the batch size, B = 64; represents the number of input channels, , H and W represent the height and width of the input feature map, ; The improved detail enhancement module (DEA-Net module) deploys 3x3 standard convolution and four differential convolutions in parallel to form five independent branches; specifically, the convolution part mainly includes standard convolution (Vanilla Conv, VC), central differential convolution (CDC), angle differential convolution (ADC), horizontal differential convolution (HDC), and vertical differential convolution (VDC); each branch uses shared Input to perform convolution operation to obtain five groups of intermediate feature maps: ; Wherein, represents the intermediate feature map output by the i-th branch, is the filter kernel (all 3x3) of the corresponding i-th branch, is the number of output channels; Subsequently, the five groups of intermediate feature maps are integrated by weighting as follows: ; Wherein, represents the weighted integration of the five branch output feature maps: first perform Softmax normalization on each branch feature in the channel dimension to obtain branch-specific weights, then weight the corresponding features with the weights and sum along the branch, thereby strengthening key details and suppressing redundant responses; represents the Softmax function; At the same time, using the additivity and linear characteristics of convolution, the five 3x3 kernels are added at the corresponding positions to obtain the equivalent kernel : ; Therefore, the entire DEA-Net module degenerates into a standard convolution in the forward inference stage, and uses a residual connection to enhance the features of the original image, and finally uses a normalization layer (BN) to normalize the output.

[0027] ; Wherein, is the final output feature map; in addition, the step is kept as 1 (stride=1), the padding is kept as 1 (padding=1), and the height and width of the output feature map are both 224, i.e. .

[0028] As shown in Figure 3 , in the present embodiment, the data processing flow of the gating bottleneck convolution module (GB_block module) is as follows: The feature map output by the C3K2 module at the last layer is taken as the input of the GB_block module :

[0029] The input feature map is reserved as a residual path:

[0030] wherein, represents an identity residual without any transformation, used for performing element-wise addition with the main branch output to alleviate gradient degradation and promote lossless transmission of low-frequency information.

[0031] The input feature map is input to the wavelet transform convolution network to improve the multi-scale frequency information of the model on image data, so as to enhance the perception ability of the model on the overall environment:

[0032] wherein, is the output feature processed by the wavelet transform convolution network on the input feature map , represents the wavelet transform convolution network; The output feature of the wavelet transform convolution network is processed by two cross-parallel data processing modules, the first data processing module is composed of a first CGT convolution module, a second CGT convolution module arranged in sequence, and the second data processing module is composed of a third CGT convolution module, a fourth CGT convolution module arranged in sequence, the outputs of the second CGT convolution module and the third CGT convolution module are used as the input of the fourth CGT convolution module after Hadamard product operation, the first CGT convolution module, the second CGT convolution module, the third CGT convolution module and the fourth CGT convolution module have the same network structure, and each includes a convolution layer (Conv), a normalization layer (GN) and a TeLU activation function (TeLU) arranged in sequence. The expression of the output data processed by the CGT convolution module is:

[0033]

[0034] wherein, represents the output data processed by the CGT convolution module, K1 is the output feature map of the wavelet transform convolution network after 1 layer of CGT convolution module, that is, the output of the third CGT convolution module, K2 is the output feature map of the wavelet transform convolution network after 2 layers of CGT convolution module, that is, the output of the second CGT convolution module.

[0035] In this embodiment, a gated fusion mechanism is adopted to perform position-channel level dynamic weighting on the input features of the fourth CGT convolution module using Hadamard product. This enables the model to adaptively adjust the information flow, strengthen important features and suppress redundant noise, thereby enhancing the small target detection capability and robustness in complex environments.

[0036] ; in, These are the input features for the fourth CGT convolutional module.

[0037] Next, the fourth CGT convolutional module processes the input features to obtain the output features of the fourth CGT convolutional module. K :

[0038] The final output is:

[0039] In this embodiment, the YOLO-Blizzard network model, which has been improved by both the DEA-Net module and the GB_block module, is used as the final target detection model. Its structure is designed to synergistically improve the target detection accuracy under harsh visual conditions such as low contrast and high noise, and solve the technical problem of insufficient feature extraction capability under blizzard weather.

[0040] Furthermore, in this embodiment, the steps for training the YOLO-Blizzard network model are as follows: Step A1: Collect vehicle images to construct the original dataset, perform preprocessing, and simultaneously complete fine-grained annotation to generate a high-quality label set; then, accurately integrate the images and labels to construct an end-to-end training dataset, providing complete and reliable input support for the YOLO-Blizzard network model; Specifically, step A1 includes the following steps: Step A11: Collect vehicle image data through multiple sources; the sources mainly include: publicly available online datasets of traffic scenes in blizzard weather, and video and image data collected offline through vehicle-mounted cameras in real road environments; the collected data covers different time periods, different weather conditions, and diverse road scenarios to ensure the broad representativeness of the dataset.

[0041] Step A12: Preprocess the acquired raw image data; the preprocessing operations mainly include: image denoising, resolution unification, format conversion and data cleaning to remove invalid and low-quality samples.

[0042] Step A13: The pre-processed image data is manually labeled using an image labeling tool (LabelImg); the labeling content is the target bounding box (Bounding Box) and the corresponding class label of each vehicle (mainly cars, buses, trucks, and trucks) in the image.

[0043] Step A14: The labeled data set is divided into a training set, a validation set, and a test set according to a predetermined ratio (8:1:1), and stored in a standard format (YOLO format) required by the target detection algorithm (YOLO series), thereby constructing a complete target detection data set that can be used for model training and evaluation.

[0044] Step A2: Train the YOLO-Blizzard network model using the training data set. Specifically, the step A2 includes the following steps: Step A21: Input the training set constructed in step A1 into the YOLO-Blizzard network model; Step A22: Set the training hyperparameters of the model; wherein the initial learning rate is set to 0.01, the optimizer is AdamW, the batch size is set to 128, and the training rounds Epochs are 500; Step A23: During the training process, the model is periodically evaluated using the validation set to monitor its performance on the validation set, mainly using the precision, recall, F1-score, mAP50, mAP75, and mAP50-95 evaluation indicators for calculation, the specific calculation formula is as follows: First, the predicted results of the model need to be compared with the true labels (Ground Truth), and the following four basic quantities are defined: TP (True Positive, True Positive): The model correctly predicts a target of a certain class, and the intersection over union (IoU) between the predicted box and the true box is greater than a predetermined threshold (such as 0.5).

[0045] FP (False Positive, False Positive): The model incorrectly predicts a target of a certain class (actually background or other classes), or the IoU between the predicted box and the true box is less than the threshold.

[0046] FN (False Negative, False Negative): The model fails to detect a target that actually exists.

[0047] TN (True Negative, True Negative): In target detection, the correct identification of background areas is usually not counted in the main evaluation, so TN has little effect in these indicators.

[0048] Precision is the measure of how many of the samples predicted as positive by the model are actually true positives. Its calculation formula is:

[0049] Recall measures how many of the true positive samples are successfully detected by the model. Its calculation formula is:

[0050] F1-Score is the harmonic mean of Precision and Recall, used to measure the overall performance of the model, especially when both need to be balanced. Its calculation formula is:

[0051] mAP (mean Average Precision) is the most core evaluation index in target detection, which comprehensively reflects the overall performance of the model in all classes and different detection thresholds. It first calculates the AP of each class, and then takes the average of the AP of all classes.

[0052]

[0053] Among them, are the average precision of the 1st, 2nd, …, nth class, and n is the total number of classes.

[0054] mAP50, mAP75, mAP50-95 are all specific variants of mAP, the difference lies in the IoU threshold used when calculating AP. The calculation formula is as follows:

[0055]

[0056]

[0057] Among them, mAP50-95 is to calculate mAP under 10 different threshold values (0.5, 0.55, 0.6…0.95) with a step of 0.05 from 0.5 to 0.95, and then take the average of the 10 mAP values. This index can best reflect the robustness of the model under different positioning accuracy requirements.

[0058] Step A24: Use the Early Stopping strategy to automatically save the best model weight file in the training process to prevent model overfitting; Step A3: save the trained model weight, select the best weight file in the training result for target detection and tracking task; Specifically, the step A3 includes the following steps: Step A31: after the model training is completed, the model weight file with the best comprehensive performance on the validation set is selected from the multiple checkpoints saved in the training process; Step A32: the selected weight file is evaluated on the independent test set to verify its generalization ability and detection performance on unseen data; Step A33: the final optimal weight file verified by the test is persistently stored as the model for subsequent target detection and tracking tasks.

[0059] In order to verify the effectiveness of the YOLO-Blizzard network model designed in the application, the application carries out an ablation experiment, and compares the designed model with other existing models, and the test results of each model are shown in Table 1.

[0060] Table 1: Comparison experiment and ablation experiment of the YOLO-Blizzard model of the application Model name mAP50 mAP75 mAP50-95 Precision Recall F1-Score YOLOv8 0.8112 0.6703 0.5857 0.8224 0.7251 0.7823 YOLOv9-t 0.8074 0.6714 0.5721 0.7839 0.7406 0.7601 YOLOv10 0.7922 0.6693 0.5773 0.8034 0.7364 0.7670 YOLOv11 0.7725 0.6177 0.5377 0.8081 0.6897 0.7430 YOLOv12 0.8016 0.6446 0.5742 0.8157 0.7664 0.7895 YOLO-DEA-Net 0.8255 0.7070 0.6070 0.8405 0.7737 0.8040 YOLO-GB_block 0.8144 0.6707 0.5858 0.8407 0.7394 0.7852 YOLO-Blizzard(ours) 0.8361 0.7015 0.6159 0.8547 0.7733 0.8110 Based on the experimental data in Table 1, it can be seen that the YOLO-Blizzard network model provided by the application exhibits good performance in the vehicle detection task facing Blizzard weather. The ablation experiment shows the contribution of the DEA-Net and the GB_block module: the DEA-Net module improves the perception ability of the model to fuzzy details in low-contrast and high-noise images by introducing a multi-path difference convolution and an adaptive weighted fusion mechanism. Its independent application (YOLO-DEA-Net) surpasses the baseline model in key indicators such as mAP50-95 and F1-Score, proving that it has good effect in enhancing the robustness of feature expression. The GB_block module uses wavelet transform and a gating mechanism to strengthen the modeling ability of the model to target morphology and environmental context, especially in improving Recall (YOLO-GB_block), which alleviates the problem of missing small targets. Both of them work together on the YOLOv11n architecture, and the finally formed YOLO-Blizzard model realizes the overall optimization of various indicators, with F1-Score improved by 6.8% compared with the baseline model, verifying the effectiveness of the module design. Therefore, DEA-Net and GB_block are not only the key components to improve detection accuracy, but also the cornerstone of stable and reliable environmental perception of the entire system in bad weather, providing input guarantee for subsequent depth estimation and risk assessment.

[0061] Embodiment 2 As Figure 4As shown, the embodiment provides a driving conflict risk assessment method for blizzard weather, which comprises the following steps: Step B1: Real-time receiving data (video stream) collected by the vehicle-mounted camera; Step B2: Using the target detection system (YOLO-Blizzard network model) for blizzard weather described in Embodiment 1 to detect the target in each frame of image; at the same time, calling the trained depth estimation convolutional network (DepthEstimationCNN) to output the depth estimation map corresponding to each frame of input image; wherein the depth estimation convolutional network (DepthEstimationCNN) is a monocular depth estimation network, which adopts an encoder-bottleneck-decoder structure.

[0062] Step B3: For each target vehicle detected around the host vehicle, based on the position of the target vehicle in the image and the corresponding depth, the real-time distance (unit: meter) of each target vehicle in the current frame is calculated and output, and the speed thereof is calculated in combination with the tracking information; Step B4: Using the OBD-II interface to access the speed data of the current vehicle (host vehicle) in real time; inputting the calculated distance and speed data of the surrounding vehicles, as well as the speed data of the host vehicle into the trained clustering risk assessment model to perform clustering analysis on the current driving state, and judging the conflict risk category in which the host vehicle is currently located in real time.

[0063] In this embodiment, the corresponding control strategy can be matched according to the risk assessment result (clustering result), and then the driver is given the corresponding conflict mitigation suggestion according to the control strategy, and the vehicle is controlled when necessary.

[0064] Further, in the embodiment, the method for calculating the speed of the target vehicle in combination with the tracking information in step B3 is: tracking the same target vehicle in consecutive video frames to obtain the spatial coordinates thereof at consecutive time points; based on the spatial displacement and time interval of the target vehicle between consecutive frames, the relative motion speed of the target vehicle relative to the host vehicle is calculated. In this embodiment, the real-time distance and relative motion speed information of the surrounding vehicles are associated and stored, which provides core data support for subsequent conflict risk assessment.

[0065] Further, in this embodiment, a lightweight depth estimation convolutional network (DepthEstimationCNN) is constructed and trained, and the model weight is saved. The complete forward propagation of the lightweight depth estimation convolutional network (DepthEstimationCNN) can be summarized as a three-stage framework of encoding-bottleneck-decoding, wherein a multi-scale cross-attention (MSCAAttention) is embedded in the bottleneck stage for multi-scale context aggregation, and the decoding stage realizes spatial resolution recovery through transposed convolution and residual fusion.

[0066] As shown in Figure 5 , the data processing flow of the lightweight depth estimation convolutional network (DepthEstimationCNN) is as follows: Given a batch of input images:

[0067] Encoder (Encoder): The encoder is composed of 4 layers of convolution-pooling units, which gradually extract abstract features and reduce the spatial resolution; among them, the first layer of convolution-pooling unit is used to extract low-level features, and the last layer of convolution-pooling unit is used to extract high-level features. The specific operation of the first layer of convolution-pooling unit is as follows:

[0068] In the formula, is a 3x3 convolution, and the number of output channels is 64, 128, 256, and 512 in turn; is batch normalization; is a self-defined activation function; The spatial size is reduced by half.

[0069] After the sequential action, the feature map size evolves as follows:

[0070]

[0071]

[0072]

[0073] Bottleneck with MSCAAttention: Multi-scale cross-attention (MSCAAttention) is applied to the lowest resolution feature map to realize multi-scale channel attention.

[0074] The multi-scale cross-attention (MSCAAttention) is internally executed as follows:

[0075]

[0076] where, represent the feature maps after normalization, represent the feature maps after normalization, .

[0077] Decoder with Skip Connections: The decoder consists of 4 layers of transpose convolution-residual units, each doubling the spatial resolution and injecting the high-resolution details preserved by the skip connections with bilinear interpolation alignment at each step. Let the k-th layer of transpose convolution-residual operation be:

[0078] where, is a 4x4 transpose convolution with stride 2; represent up-sampling with bilinear interpolation to restore the size, represent the feature maps obtained in the encoding stage, represent the normalization of each layer.

[0079] After layer-by-layer execution, the feature map size evolves as follows: ; ; ; ; Finally, for the output of the decoder, a 1x1 convolution is used to reduce the number of channels to 1, and then bilinear interpolation is used to obtain the specified output size, the expression is: ;

[0080] where, indicates that the spatial size is interpolated to 55x74, represent the 1x1 convolution to compress 64-channel features to 1 channel, realizing cross-channel information fusion and reducing dimension.

[0081] In summary, the Depth Estimation CNN extracts semantics through step-by-step down-sampling, aggregates multi-scale context at the bottleneck using multi-scale cross-attention (MSCA Attention), and then recovers spatial details through a decoder with skip connections, finally outputs a depth estimation map that corresponds one-to-one with the input.

[0082] Further, as Figure 6 ​As shown, in this embodiment, the core of the MSCAAttention attention mechanism lies in its dual-path parallel structure, aiming to perform multi-scale context modeling from two orthogonal dimensions (horizontal and vertical) of space Specifically, this module first sends the input features into two independent processing branches: first, after standardization, the upper branch passes through the parallel application of 1x7, 1x11 and 1x21 depth convolution kernels to extract different scale horizontal spatial dependencies, and the lower branch passes through the parallel application of 7x1, 11x1 and 21x1 depth convolution kernels to capture multi-scale vertical spatial dependencies; second, within each branch, the output is subjected to 1x1 convolution for cross-channel information interaction and integration, and query (Q), key (K) and value (V) matrices are generated through linear transformation. Then, the Q matrix generated by the upper branch (Fx) and the Q matrix generated by the lower branch (Fy) are exchanged, and the exchanged Q matrices are multiplied by the corresponding K matrices, and then multiplied by the V matrix to generate the attention feature maps of the upper and lower branches respectively. Next, the attention feature maps of the two branches are reshaped and mixed through 1x1 convolution. Finally, the outputs Fup of the upper branch and Fdown of the lower branch are fused to realize global spatial context information enhancement, thereby improving the perception accuracy of object boundaries and structural details.

[0083] In this embodiment, an image dataset with real depth labels (obtained by laser radar) is collected or constructed as training data for a depth estimation convolutional network (DepthEstimationCNN); the depth estimation convolutional network (DepthEstimationCNN) is trained end-to-end using this dataset to optimize its depth estimation capability, and the main steps of its training process are as follows: Considering that the ground truth depth map (ground truth) is very sparse (obtained by LiDAR scanning, with few points in the distance), direct gradient calculation will lead to inaccuracy, which will interfere with the convergence of data loss, so the present application adopts a combined loss function Lt, which consists of two main parts: data loss (DataLoss, Ld) and gradient loss (GradientLoss, Lg), and the specific formula is as follows: ;

[0084] Wherein, : is the total loss function that the network needs to minimize during training; : the predicted depth map (Predicted depth map); : the real depth map (Ground truth depth map); : data loss, measures the overall difference between predicted depth values and ground truth depth values; : gradient loss, measures the difference between predicted depth map and ground truth depth map in edge (gradient); a and b: these are two balancing factors, used to adjust and the weights in total loss.

[0085] The expression of data loss is: ;

[0086] This loss function is used to measure the overall difference between predicted depth map and ground truth depth map . Wherein, is the difference between the predicted depth and the ground truth depth of the th valid pixel in the logarithmic space, using the logarithmic space is to deal with the imbalance of depth data: the depth value of the near distance is dense and changes greatly, the depth value of the far distance is sparse and changes little, and the logarithmic transformation can alleviate this imbalance; is the sum of squares of all valid pixels , representing the overall size of the prediction error; is the square of the sum of all , representing the systematic deviation of the prediction error. is the set of valid pixels in the depth map (i.e. pixels with ground truth depth values). is the total number of valid pixels; is a balancing factor, set to 0.85 in this embodiment, which controls the degree of punishment of the loss function to the systematic deviation.

[0087] The expression of gradient loss is: ; The gradient loss function is specifically used to enhance the local details of the depth map, especially the depth boundary. It is achieved by minimizing the gradient difference between the predicted depth map and the ground truth depth map in the horizontal and vertical directions. Wherein, and are the gradient values of the predicted depth map in the horizontal direction and the vertical direction of the th pixel. represents the true value map, since the original ground truth depth map is very sparse, directly calculating its gradients and will get a lot of invalid values. Therefore, in this embodiment, the interpolation tool kit is used to interpolate the sparse true value to obtain a dense true value map which can be used to calculate the gradient; is the L1 norm (absolute value) difference of horizontal direction gradient between the predicted depth map and the interpolated ground truth map. is the L1 norm difference of vertical direction gradient.

[0088] During the training process, the application also monitors the depth estimation error (RMSE, MSE, etc.) on the validation set, and saves the model weight file with the best performance for subsequent distance estimation tasks.

[0089] In order to verify the effectiveness of the depth estimation convolutional network designed by the application, the application compares it with other existing models, and the test results of each model are shown in Table 2.

[0090] Table 2: Comparison of depth estimation model experimental results Model MAE↓ RMSE↓ REL↓ Total parameter amount Parameter size(MB) CNN 2.6723 2.9811 0.9726 4,307,329 16.43 InceptionV3 2.6688 2.977 0.9714 4,596,001 17.53 ResNet 2.6509 2.9679 0.9608 32,553,921 127.18 MobileNetV2 2.6438 2.9563 0.9595 15,465,601 59.00 The invention(Ours) 2.6394 2.9556 0.9561 4,626,305 17.65 Based on the experimental data in Table 2, it can be seen that the DepthEstimationCNN proposed in the application realizes good lightweight while ensuring high precision. Compared with mainstream models, the MAE, RMSE and REL three core error indicators are all optimal, especially in the REL indicator sensitive to depth boundaries, which proves its ability to restore object contours and spatial relationships in snowstorm scenes. At the same time, the parameter amount of the model is only about 4.6M, which is much lower than ResNet (32.5M) and MobileNetV2 (15.5M), and the parameter size (17.65MB) is comparable to the basic CNN model, which embodies the "lightweight" design advantage. This makes it possible to deploy on a vehicle embedded platform, providing accurate and reliable depth input for subsequent conflict risk assessment under the premise of meeting real-time requirements, which is the technical support for the realization of closed-loop perception and early warning of the entire system.

[0091] Further, in the embodiment, the construction method of the clustering risk assessment model is: Step C1: taking the multiple sets of target vehicle distance and relative motion speed data obtained in step B3 and the speed of the ego vehicle (own vehicle) as input features, a feature vector for traffic conflict risk assessment is constructed; Step C2: using an unsupervised clustering algorithm (K-means clustering algorithm) to perform clustering analysis on the feature vector; Step C3: defining different risk level categories by analyzing the clustering results, and classifying and describing according to the clustering centers.

[0092] In the embodiment, it is clustered into six categories: the first category is "low-speed long-distance safe cruise class", which is characterized by the average distance of various surrounding vehicles being far, the average speed being low or similar to the ego vehicle, the ego vehicle also being in a stable driving state, the overall traffic situation being relaxed, and the risk level being low; the second category is "medium-speed medium-distance stable following class", many types of surrounding vehicles are in a medium-distance range, the speed and the ego vehicle are highly matched, an orderly traffic flow is formed, the ego vehicle speed is moderate, and the risk level is low to medium; the third category is "high-speed short-distance high-risk approach class", at least one type of surrounding vehicle is close and has a high relative speed, or the ego vehicle is driving at high speed in a close-range vehicle group, there is a significant risk of rear-end or lateral collision, and the risk level is high; the fourth category is "low-speed congestion following class", the average speed of various surrounding vehicles is low, the spacing is small, the ego vehicle is in a low-speed following state, and it is often seen in traffic congestion scenes, although the relative speed is low but the collision probability is high, and the risk level is medium; the fifth category is "high-risk interference class of different types of vehicles", specific types of vehicles (such as large trucks or buses) show abnormal speed changes or extremely close approach, because of the large difference in their kinetic characteristics, they are easy to cause sudden conflicts, and the risk level is high; the sixth category is "high-speed free-flow cruise class", the surrounding vehicles are distributed sparsely and far apart, each type of vehicle and the ego vehicle maintains a high speed and stable driving, the traffic flow is smooth, and the risk level is low. The above six categories cover typical driving situations in snowstorms, providing reliable semantic risk criteria for the development of differentiated control strategies.

[0093] In this embodiment, the six types of driving scenes identified by the clustering risk assessment model clustering analysis are differentiated according to their risk characteristics to formulate differentiated conflict resolution strategies. For the "low-speed long-distance safe cruise class" and the "high-speed free-flow cruise class", the overall traffic situation is relaxed, the distance between vehicles is sufficient and the relative speed is stable, it is determined as low risk, the system only performs regular state monitoring, does not trigger active intervention, to ensure the smoothness and comfort of driving; for the "medium-speed medium-distance stable following class", the vehicle is in normal following state, the system suggests maintaining the current vehicle distance and continuously monitoring the front vehicle dynamics, and if necessary, prompting the driver to adjust the speed to cope with potential speed fluctuations; for the "low-speed congestion following class", although the vehicle speed is low, the distance between vehicles is small and the density is high, and it is easy to cause a chain collision due to the delay of the driver's reaction, so the system starts the automatic following control (ACC) and suggests smooth braking and maintaining a safe buffer to reduce the risk of rear-end collision; for the "high-speed close-distance high-risk approaching class", since the ego vehicle or surrounding vehicles are in a high-speed state and the distance is too close, the braking distance is insufficient, the system immediately triggers high-level audible and visual warnings, and suggests emergency deceleration or lane change to avoid, and if necessary, performs pre-braking or emergency braking assistance through the vehicle control interface; for the "high-risk interference of different types of vehicles", the system focuses on monitoring the abnormal motion behavior of large vehicles (such as trucks and buses), because of their large mass and brake hysteresis, the system sends a local risk warning and suggests actively increasing the lateral and longitudinal safety distance to avoid being in their blind area or following them closely to avoid sudden danger.

[0094] In this embodiment, the trained clustering risk assessment model and its classification rules and response strategies are solidified and saved to form an evaluation model that can be used for real-time risk discrimination.

[0095] The application also provides an electronic device, comprising: one or more processors, a memory; wherein the memory is used to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the driving conflict risk assessment method for blizzard weather described above.

[0096] The application also provides a computer readable medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the driving conflict risk assessment method for blizzard weather described above.

[0097] Those skilled in the art can understand that all or part of the functions of various methods / modules in the above embodiments can be implemented by hardware or by a computer program. When all or part of the functions are implemented by a computer program, the program can be stored in a computer readable storage medium, which can include read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc. The above functions are implemented by executing the program by a computer. For example, the program is stored in the memory of the device, and the above functions are implemented by executing the program in the memory by the processor.

[0098] In addition, when all or part of the functions in the above embodiments are implemented by a computer program, the program can also be stored in a storage medium such as a server, another computer, a disk, an optical disk, a flash disk, or a mobile hard disk, and saved in the memory of the local device by downloading or copying, or the system of the local device is updated, and the above functions are implemented by executing the program in the memory by the processor.

[0099] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope described in the claims.

Claims

1. A target detection system for inclement weather, characterized by, The target detection system is obtained by training a YOLO-Blizzard network model, the YOLO-Blizzard network model is based on a YOLOv11n model as a basic network model, an improved detail enhancement module is embedded between a first Conv module and a second Conv module of a backbone network of the YOLOv11n model, and a convolution module in a C3K2 module of a neck network of the YOLOv11n model is replaced into a gated bottleneck convolution module; The improved detail enhancement module comprises one standard convolution and four differential convolutions, the input data is first extracted by five parallel independent branches, and then the extracted features are weighted and integrated; then the filter kernels of the five independent branches are added to obtain an equivalent kernel, and the equivalent kernel and the features obtained by the weighted integration are convolved again; finally, a residual connection is introduced and processed through a normalization layer; The gated bottleneck convolution module first captures multi-scale frequency information of an image by using a wavelet transform convolution network, the output features of the wavelet transform convolution network are processed by two cross-parallel data processing modules, the first data processing module is composed of a first CGT convolution module, a second CGT convolution module arranged in sequence, the second data processing module is composed of a third CGT convolution module, a fourth CGT convolution module arranged in sequence, the outputs of the second CGT convolution module and the third CGT convolution module are subjected to Hadamard product operation to obtain the input of the fourth CGT convolution module, then the output features of the fourth CGT convolution module are subjected to element-wise addition operation with the output features of the wavelet transform convolution network to complete feature fusion; finally, the fused features are subjected to nonlinear transformation, and a residual connection is performed with the original input features to obtain the final output.

2. The storm weather oriented target detection system of claim 1, wherein, The four differential convolutions in the improved detail enhancement module are central differential convolution, angle differential convolution, horizontal differential convolution and vertical differential convolution; when the extracted features are weighted and integrated, the branch-specific weight is obtained by performing Softmax normalization on the channel dimension of each branch feature, and the corresponding features are weighted and summed along the branch.

3. The inclement weather oriented target detection system of claim 1, wherein, The first CGT convolution module, the second CGT convolution module, the third CGT convolution module and the fourth CGT convolution module have the same network structure and comprise a convolution layer, a normalization layer and a TeLU activation function arranged in sequence.

4. A driving conflict risk assessment method for adverse weather, characterized by, The method comprises the following steps: Step B1: real-time receiving data collected by a vehicle-mounted camera; Step B2: using the target detection system for blizzard weather in any one of claims 1 to 3 to detect targets in each frame of image; at the same time, calling a trained depth estimation convolution network to output a depth estimation map corresponding to each input frame of image; wherein the depth estimation convolution network is a monocular depth estimation network, which adopts an encoder-bottleneck-decoder structure, embeds multi-scale cross attention in the bottleneck stage for multi-scale context aggregation, and realizes spatial resolution recovery through transposed convolution and residual fusion in the decoding stage; Step B3: For each target vehicle detected around the host vehicle, the real-time distance of each target vehicle in the current frame is calculated and output based on the position of the image where the target vehicle is located and the corresponding depth, while the speed of each target vehicle is calculated in combination with the tracking information; Step B4: Real-time access to the speed data of the host vehicle through the OBD-II interface; input the calculated distance and speed data of the surrounding vehicles, as well as the speed data of the host vehicle into the trained clustering risk assessment model to perform clustering analysis on the current driving state and judge the conflict risk category of the host vehicle in real time.

5. The driving conflict risk assessment method for severe weather, according to claim 4, wherein, The method for calculating the speed of the target vehicle in combination with the tracking information in step B3 is: tracking the same target vehicle in consecutive video frames to obtain the spatial coordinates of the target vehicle at consecutive time points; based on the spatial displacement and time interval of the target vehicle between consecutive frames, the relative motion speed of the target vehicle relative to the host vehicle is calculated.

6. The driving conflict risk assessment method for severe weather, according to claim 4, wherein, The encoder in the depth estimation convolutional network consists of 4 layers of convolution-pooling units, which extract abstract features and reduce spatial resolution step by step; wherein, the specific operation of the first layer of convolution-pooling unit is as follows: The specific operation of the first layer of convolution-pooling unit is as follows: ； In the formula, is a 3x3 convolution, and the number of output channels is 64, 128, 256, and 512 in turn; is batch normalization; is a self-defined activation function; The spatial size is halved.

7. The driving conflict risk assessment method for adverse weather conditions according to claim 6, characterized in that, The decoder in the depth estimation convolutional network is composed of 4 layers of transpose convolution-residual units, each step doubles the spatial resolution, and the high-resolution details preserved in the encoding stage are injected through the bilinear interpolation aligned skip connection, and the kth layer of transpose convolution-residual operation is: ； wherein, is a 4x4 transposed convolution with stride 2; represents an up-sampling with bilinear interpolation to recover the size, represents the features obtained at the encoding stage, represents the normalization of each layer.

8. The driving conflict risk assessment method for severe weather, according to claim 7, wherein, The multi-scale cross-attention embedded in the bottleneck stage adopts a double-path parallel structure, and the input features are sent into two independent processing branches: first, after standardization, the upper branch is applied with 1x7, 1x11 and 1x21 depth convolution kernels in parallel to extract different scale horizontal spatial dependencies, and the lower branch is applied with 7x1, 11x1 and 21x1 depth convolution kernels in parallel after standardization to capture multi-scale vertical spatial dependencies; secondly, within each branch, the output is subjected to 1x1 convolution for cross-channel information interaction and integration, and the query Q, key K and value V matrices are generated through linear transformation; then, the Q matrix generated by the upper branch and the Q matrix generated by the lower branch are exchanged, and the exchanged Q matrix is multiplied by the corresponding K matrix, and then multiplied by the V matrix to generate the attention feature map of each branch; then, the attention feature maps of the two branches are reshaped and mixed through 1x1 convolution; finally, the outputs Fup of the upper branch and Fdown of the lower branch are fused to enhance the global spatial context information.

9. The driving conflict risk assessment method for adverse weather conditions according to claim 8, characterized in that, When training the depth estimation convolutional network, the total loss function is composed of data loss and gradient loss, and the expression is: ； wherein, is the total loss function that the network needs to minimize during the training process; : predicted depth map by the network; : ground truth depth map; : data loss, measuring the overall difference in numerical values between the predicted depth values and the ground truth depth values; : gradient loss, measuring the difference in edges between the predicted depth map and the ground truth depth map; a and β are two balancing factors to adjust the weights of and in the total loss.

10. The driving conflict risk assessment method for severe weather conditions according to claim 9, characterized in that, The expression of the data loss is: ； wherein, is the difference between the predicted depth and the true depth in log space for the first valid pixel, is the sum of the squares of the differences between the predicted depth and the true depth in log space for all valid pixels, is the square of the sum of the differences between the predicted depth and the true depth in log space for all valid pixels, is the set of valid pixels in the depth map, is the total number of valid pixels; is a balancing factor;​​ The expression of the gradient loss is: ； in, and These are the predicted depth maps. In the The gradient values ​​in the horizontal and vertical directions at each pixel; Represents the truth graph; It represents the difference in the L1 norm of the horizontal gradient between the predicted depth map and the interpolated ground truth map. It is the L1 norm difference on the vertical gradient.

Images