Foggy day target detection method and device based on YOLOv12 optimization
By improving the YOLOv12 model, adding the defog module and joint training loss function, the problem of insufficient target detection accuracy in fog-day environments is solved, and more efficient target detection in fog-days is achieved.
Patent Information
- Application Number
- CN202510547723.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-08-29
AI Technical Summary
The existing algorithms such as YOLOv5 and YOLOv8 cannot meet the target detection accuracy requirements in harsh environments such as foggy days, resulting in a decrease in detection accuracy and missed detection and missed detection.
By improving the YOLOv12 model, the defog removal module and joint training loss function are added, combined with the feature extraction module and the ConvTranspose block, the dynamic loss weight strategy is used for training to realize the joint optimization of image defog removal and object detection.
It improves the accuracy and efficiency of target detection in foggy days, and can more accurately identify and locate traffic targets in foggy days.
Smart Images
Figure CN120564142A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of target detection, and in particular to a method and device for foggy target detection based on YOLOv12 optimization. Background Art
[0002] Object detection, a key problem in computer vision, aims to identify and locate all objects of interest in an image or video, while also determining their categories. While existing object detection methods achieve good results under normal lighting conditions, they often suffer from reduced image brightness and contrast in adverse weather conditions, such as fog, haze, or low-light environments. This can severely impact recognition, leading to decreased object detection accuracy and the occurrence of missed and false detections.
[0003] To address the impact of inclement weather on algorithm accuracy, recent research has focused on improving object detection performance in adverse weather conditions. Currently, object detection algorithms are primarily categorized into two types: two-stage object detection algorithms, such as Fast R-CNN, Faster R-CNN, and Mask R-CNN; and one-stage object detection algorithms, such as SSD, YOLO, YOLOv5, and YOLO v8. Compared to contemporary two-stage detection algorithms, one-stage detection algorithms offer lower accuracy and higher speed. Due to their strong real-time performance, they are widely used in practical engineering applications.
[0004] However, with the passage of time and technological advancements, commonly used algorithms such as YOLOv5 and YOLOv8 in existing technologies can no longer meet the accuracy requirements for object detection in harsh environments such as foggy days. Therefore, it is necessary to further improve algorithm performance and thereby enhance the accuracy and efficiency of foggy object detection. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a foggy target detection method and device based on YOLOv12 optimization, which can improve the accuracy and efficiency of foggy target detection.
[0006] An embodiment of the present invention provides a method for detecting targets in foggy weather based on YOLOv12 optimization, comprising the following steps:
[0007] Collecting raw data and preprocessing the raw data to obtain training data; wherein the training data includes at least a plurality of traffic image data in a foggy environment;
[0008] According to the training data, a preset improved YOLOv12 model is trained to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function;
[0009] The foggy weather target detection model is used to perform target detection on the data to be detected to obtain a detection result.
[0010] Furthermore, the collecting of raw data and preprocessing of the raw data to obtain training data specifically includes:
[0011] Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments;
[0012] The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object width and object height of all traffic objects in the original data.
[0013] Preferably, after preprocessing the raw data, the method further comprises:
[0014] Simulated foggy environment data is generated by a preset image simulation algorithm, and the simulated foggy environment data is added to the training data.
[0015] Furthermore, the construction process of the preset improved YOLOv12 model specifically includes:
[0016] The defogging module is added after the input end of the preset YOLOv12 model, and a feature extraction module is added to the network layer of the preset YOLOv12 model. At the same time, the loss function of the preset YOLOv12 model is replaced by the joint training loss function, and the Upsamle block in the preset YOLOv12 model is replaced by the ConvTranspose block, finally obtaining the preset improved YOLOv12 model; wherein, the defogging module is composed of several Conv modules and DSconv modules, and is used to defog the traffic image data in the foggy environment to generate a clear image.
[0017] Preferably, the joint training loss function specifically includes an object detection loss function and an image defogging and restoration loss function, and the expression of the joint training loss function is specifically:
[0018] Loss=α*DatectLoss+β*DehazeLoss
[0019] Among them, Loss is the joint loss value, DetectLoss is the object detection loss function, DehazeLoss is the image dehazing and restoration loss function, α is the detection weight corresponding to the object detection loss function, and β is the dehazing weight corresponding to the image dehazing and restoration loss function.
[0020] Furthermore, the preset improved YOLOv12 model is trained based on the training data to obtain a foggy target detection model, specifically including:
[0021] Dividing the training data into a training set, a validation set, and a test set by stratified sampling;
[0022] According to the training set, validation set and test set, the preset improved YOLOv12 model is trained through iterative training until the joint loss value is less than the preset loss threshold, and the training is terminated to obtain the foggy target detection model.
[0023] Preferably, the training process further includes: setting a dynamic loss weight strategy; wherein the dynamic loss weight strategy is specifically:
[0024] During the iterative training process, the detection weight and the defogging weight are set to linearly decrease and linearly increase respectively.
[0025] Furthermore, the target detection is performed on the data to be detected by the foggy target detection model to obtain a detection result, which specifically includes:
[0026] The data to be detected is input into the foggy weather target detection model for defogging and target detection to obtain the detection result; wherein, the detection result includes a clear image corresponding to the data to be detected and the category and prediction frame information of the traffic object contained in the data to be detected, and the prediction frame information includes the center coordinates and the width and height of the traffic object.
[0027] Another embodiment of the present invention provides a foggy target detection device based on YOLOv12 optimization, including: an acquisition module, a training module, and a detection module;
[0028] The acquisition module is used to acquire raw data and pre-process the raw data to obtain training data; wherein the training data at least includes a plurality of traffic image data in a foggy environment;
[0029] The training module is used to train a preset improved YOLOv12 model based on the training data to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function;
[0030] The detection module is used to perform target detection on the data to be detected by using the foggy weather target detection model to obtain a detection result.
[0031] Furthermore, the acquisition module is used to collect raw data and pre-process the raw data to obtain training data, specifically including:
[0032] Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments;
[0033] The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object width and height of all traffic objects in the original data.
[0034] Compared with the prior art, the present invention has the following beneficial effects:
[0035] By improving the original YOLOv12 model and adding a defogging module, the improved model is more adaptable to traffic target detection in foggy environments. Furthermore, by setting a dynamic joint training loss function, image defogging and traffic target detection tasks can be jointly trained, thereby improving the overall performance of the model. In summary, the YOLOv12-optimized foggy target detection method provided by this invention can improve the accuracy and efficiency of foggy target detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 A flowchart of a method for foggy target detection based on YOLOv12 optimization is provided in accordance with one embodiment of the present invention.
[0037] Figure 2 A schematic diagram of the structure of a preset improved YOLOv12 model provided by one embodiment of the present invention.
[0038] Figure 3 A schematic structural diagram of a defogging module provided in one embodiment of the present invention.
[0039] Figure 4 A schematic structural diagram of a DSconv module provided in one embodiment of the present invention.
[0040] Figure 5 A schematic diagram of the structure of a foggy target detection device based on YOLOv12 optimization is provided in another embodiment of the present invention. DETAILED DESCRIPTION
[0041] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;
[0042] It is understandable to those skilled in the art that some well-known structures and descriptions thereof may be omitted in the drawings.
[0043] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.
[0044] Reference Figure 1 , which is a flowchart of a method for foggy target detection based on YOLOv12 optimization provided by one embodiment of the present invention, including the following steps:
[0045] S1: collecting raw data and preprocessing the raw data to obtain training data; wherein the training data includes at least a plurality of traffic image data in a foggy environment;
[0046] S2: Training a preset improved YOLOv12 model based on the training data to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function;
[0047] S3: Performing target detection on the data to be detected using the foggy weather target detection model to obtain a detection result.
[0048] Regarding step S1, specifically, collecting raw data and preprocessing the raw data to obtain training data specifically includes:
[0049] Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments;
[0050] The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object width and height of all traffic objects in the original data.
[0051] Preferably, after preprocessing the raw data, the method further comprises:
[0052] Simulated foggy environment data is generated by a preset image simulation algorithm, and the simulated foggy environment data is added to the training data.
[0053] In a preferred embodiment, raw data of various environmental traffic backgrounds is obtained by collecting data from public databases, web crawlers, etc., and labeling is completed for the raw data. The labeling method uses the labelimg tool to obtain the labeling information. The labeling information is in the form of: each image corresponds to a txt file, and the txt file contains multiple lines of information, where each line of information includes: the type of traffic object in the image, the x and y coordinates of the center of the traffic object, and the length and width information of the traffic object frame, where the length and width information is: the ratio of the length and width of the traffic object frame to the length and width of the image. One line represents the information of one object, so the number of lines corresponds to the number of traffic objects in the image.
[0054] Considering that traffic image data under foggy conditions is relatively scarce, after acquiring the original data, an algorithm can be used to simulate and generate some traffic image data under foggy conditions. The generated dataset is merged with the original data to obtain complete training data.
[0055] For step S2, specifically, the construction process of the preset improved YOLOv12 model specifically includes:
[0056] The defogging module is added after the input end of the preset YOLOv12 model, and a feature extraction module is added to the network layer of the preset YOLOv12 model. At the same time, the loss function of the preset YOLOv12 model is replaced by the joint training loss function, and the Upsamle block in the preset YOLOv12 model is replaced by the ConvTranspose block, finally obtaining the preset improved YOLOv12 model; wherein, the defogging module is composed of several Conv modules and DSconv modules, and is used to defog the traffic image data in the foggy environment to generate a clear image.
[0057] Preferably, the joint training loss function specifically includes an object detection loss function and an image defogging and restoration loss function, and the expression of the joint training loss function is specifically:
[0058] Loss=α*DetectLoss+β*DehazeLoss
[0059] Among them, Loss is the joint loss value, DetectLoss is the object detection loss function, DehazeLoss is the image dehazing and restoration loss function, α is the detection weight corresponding to the object detection loss function, and β is the dehazing weight corresponding to the image dehazing and restoration loss function.
[0060] In a preferred embodiment, referring to Figure 2 , is a schematic diagram of the structure of a preset improved YOLOv12 model provided by an embodiment of the present invention. Figure 2It can be seen that compared with the original YOLOv12 model, the preset improved YOLOv12 model provided in this preferred embodiment adds a defogging module and a feature extraction module, sets a joint training loss function, and finally replaces the original upsampling module with a ConvTranspose block.
[0061] Since the original YOLOv12 lacks image defogging function, a defogging module is added for foggy environments. Figure 3 , is a schematic diagram of the structure of a defogging module provided by an embodiment of the present invention. Figure 3 It can be seen that the defogging module consists of several Conv modules and DSconv modules.
[0062] When the original fogged image is input, it passes through the stacked Conv and DSconv modules for feature extraction. A richer feature representation is obtained by merging shallow and deep features. This feature is primarily responsible for extracting abstract texture features from the fogged image. Finally, the abstract features are added point by point to the original image to obtain the dehazed image. The dehazed image is compared with the corresponding clear image in the training set, and the dehazing loss function is calculated. Backpropagation of the loss function is used to update the parameters of the dehazing module, ultimately enabling the module to produce a dehazed image that approximates the clear image, thus obtaining the learned parameters. These parameters are solidified during testing or use, allowing the generation of a clear image from the fogged image. The cleared image is then input into the subsequent detection network.
[0063] Reference Figure 4 , is a structural diagram of a DSconv module provided by an embodiment of the present invention. Figure 4 As can be seen, the input feature map first passes through the DWconv depthwise separable convolution module for feature extraction. DWconv uses large 9*9 or 7*7 convolution kernels to increase the receptive field and reduces the number of parameters and computation through channel-wise separable convolution. BatchNorm is then used for normalization, improving the stability of model training and accelerating model convergence. The Conv module then performs cross-channel learning and channel transformation, and the Softplus activation function performs a nonlinear transformation on the feature map, improving the module's nonlinear representation and feature fitting capabilities. Finally, the feature map is added to the input feature map point by point, and the resulting feature map is output.
[0064] While adding a dehazing module, the problem of large object sizes and a large variance in the number of traffic objects present during detection was considered. Therefore, this preferred embodiment also adds a feature extraction module to the YOLOv12 model's network layer, increasing the network depth and adding corresponding detection heads to YOLOv12. The modified YOLOv12 includes a total of four detection heads of different sizes.
[0065] In addition, this preferred embodiment also proposes an image defogging and restoration loss function, thereby modifying the original YOLOv12 model loss function to the joint training loss function, thereby achieving joint training and learning of the object detection and defogging functions of the YOLOv12 model. The joint training loss function includes the object detection loss function and the image defogging and restoration loss function. The loss functions of the two tasks are weighted and calculated using the following formula:
[0066] Loss=α*DetectLoss+β*DehazeLoss
[0067] Among them, DetectLoss consists of the category loss function BCELoss, the prediction box position loss function CIoULoss and the prediction box position distribution loss function DFLoss.
[0068] The specific formula of the category loss function BCELoss is:
[0069]
[0070] where p i (c) represents the category label of object i, Represents the category prediction result of object i.
[0071] The specific formula of the prediction box position loss function CIoULoss is:
[0072] CIoULoss=1-CIoU
[0073] Among them, CIoU represents the intersection-over-union ratio of the predicted box and the label box.
[0074] The specific formula of the prediction box position distribution loss function DFLoss is:
[0075] DFLoss(S i ,S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 ))
[0076] Among them, y i and y i+1 Represents the left and right nearest neighbor values of the label value y, S i and S i+1 Respectively represent the label value y in y i and y i+1 probability.
[0077] DehazeLoss is composed of the dehazing pixel loss function Dehaze_ContentLoss and the dehazing feature loss functions Dehaze_FeatureLoss and Dehaze_SSIM. The specific formulas of the three are as follows:
[0078]
[0079] Among them, ∈ represents the minimum value close to 0, c and Represent the true value and reconstructed value of the pixel after defogging. f(c) and They represent the feature values of the dehazed ground truth and reconstructed values after mapping through a pre-trained backbone model. The pre-trained backbone model can be a classic model such as VGG, ResNet, and DenseNet. Through loss supervision at both the pixel and feature levels, the dehazing module can be better trained.
[0080] Finally, this preferred embodiment also changes the original upsampling module of the YOLOv12 model from Upsamle to the deconvolution ConvTranspose method. This is because upsampling is a relatively mechanical method of resampling the nearby neighborhood. In order to allow the model to obtain more flexible and efficient upsampled feature maps, this preferred embodiment chooses to use deconvolution operations to allow the model to automatically learn, thereby enabling the model to adapt to various actual situations. Merging the feature maps after ConvTranspose with the shallow feature maps can improve feature diversity, thereby achieving more effective object detection.
[0081] Regarding step S2, further, the preset improved YOLOv12 model is trained according to the training data to obtain a foggy target detection model, specifically including:
[0082] Dividing the training data into a training set, a validation set, and a test set by stratified sampling;
[0083] According to the training set, validation set and test set, the preset improved YOLOv12 model is trained through iterative training until the joint loss value is less than the preset loss threshold, and the training is terminated to obtain the foggy target detection model.
[0084] Preferably, the training process further includes: setting a dynamic loss weight strategy; wherein the dynamic loss weight strategy is specifically:
[0085] During the iterative training process, the detection weight and the defogging weight are set to linearly decrease and linearly increase respectively.
[0086] In a preferred embodiment, a dynamic loss weighting strategy is used to stabilize the joint training of the dehazing and detection modules. In this preferred embodiment, the loss weights α and β are set to decrease and increase linearly, respectively. During iterative training, α decreases linearly from 1 to 0.5, while β increases linearly from 0.5 to 1. This setting allows the early stages of model training to prioritize learning the image dehazing function, while the later stages prioritize learning the detection function.
[0087] In actual training, the specific training steps are as follows:
[0088] (1) Load the training data into the model for training. Divide the training data into 50% as a training set, 30% as a validation set, and 20% as a test set. During the data partitioning process, stratified sampling is performed on the datasets from different sources to ensure similar data distribution.
[0089] (2) Build the environment required for the preset improved YOLOv12 model training, such as installing Pytorch, Opencv and other libraries, configuring the GPU environment, etc.
[0090] (3) Set the hyperparameters for network model training. Some hyperparameters are set to: Epochs: 500, Optimizer: SGD, Momentum: 0.95, Batch size: 16, Weight decay: 0.0001, Warm-up epochs: 2, Warm-up momentum: 0.8, Initial learning rate: 0.005, Final learning rate: 0.0002, Learning rate schedule: Linear decay.
[0091] (4) Iterative training is set until the joint loss value is less than a preset loss threshold (the specific value can be determined according to the actual situation), and then the training is terminated and the foggy target detection model is output.
[0092] Regarding step S3, specifically, performing target detection on the data to be detected by using the foggy weather target detection model to obtain a detection result specifically includes:
[0093] The data to be detected is input into the foggy weather target detection model for defogging and target detection to obtain the detection result; wherein, the detection result includes a clear image corresponding to the data to be detected and the category and prediction frame information of the traffic object contained in the data to be detected, and the prediction frame information includes the center coordinates and the width and height of the traffic object.
[0094] In summary, the method for foggy weather target detection based on YOLOv12 optimization provided by the present invention first completes the initial dataset annotation for a variety of environmental traffic scene image datasets. Then, based on the data enhancement processing of the original dataset, data expansion is performed through techniques such as image light processing and background environment processing to obtain a new dataset. Since there is less traffic image data in foggy environments, an algorithm is used to generate some traffic image data in foggy environments. The generated dataset and the original foggy dataset are merged to train the model together. Then, a defogging module is added to the YOLOv12 model to adapt it to traffic object detection in foggy environments. In addition, the feature extraction module and detection head of the model are optimized to further improve its recognition ability. Then, the improved YOLOv12 model is used for data training and learning to obtain a foggy weather traffic target detection model, which is finally applied in intelligent traffic recognition. Compared with existing target detection models, the accuracy and efficiency of foggy weather target detection can be improved.
[0095] Reference Figure 5 , is a structural diagram of a foggy target detection device based on YOLOv12 optimization provided by another embodiment of the present invention, including: an acquisition module 101, a training module 102, and a detection module 103;
[0096] The acquisition module 101 is used to acquire raw data and pre-process the raw data to obtain training data; wherein the training data at least includes a plurality of traffic image data in a foggy environment;
[0097] The training module 102 is used to train a preset improved YOLOv12 model based on the training data to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function;
[0098] The detection module 103 is used to perform target detection on the data to be detected by using the foggy weather target detection model to obtain a detection result.
[0099] Furthermore, the acquisition module 101 is used to acquire raw data and pre-process the raw data to obtain training data, specifically including:
[0100] Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments;
[0101] The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object width and height of all traffic objects in the original data.
[0102] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.
Claims
1. A foggy target detection method based on YOLOv12 optimization, characterized in that: The steps include: Collecting raw data and preprocessing the raw data to obtain training data; wherein the training data includes at least a plurality of traffic image data in a foggy environment; According to the training data, a preset improved YOLOv12 model is trained to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function; The foggy weather target detection model is used to perform target detection on the data to be detected to obtain a detection result.
2. The foggy target detection method based on YOLOv12 optimization according to claim 1, characterized in that: The collecting of raw data and preprocessing of the raw data to obtain training data specifically includes: Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments; The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object width and object height of all traffic objects in the original data.
3. The foggy target detection method based on YOLOv12 optimization according to claim 2, characterized in that: After preprocessing the raw data, the method further includes: Generate simulated foggy environment data, and add the simulated foggy environment data to the training data.
4. The foggy target detection method based on YOLOv12 optimization according to claim 1, characterized in that The construction process of the preset improved YOLOv12 model specifically includes: The defogging module is added after the input end of the preset YOLOv12 model, and a feature extraction module is added to the network layer of the preset YOLOv12 model. At the same time, the loss function of the preset YOLOv12 model is replaced by the joint training loss function, and the Upsamle block in the preset YOLOv12 model is replaced by the ConvTranspose block, finally obtaining the preset improved YOLOv12 model; wherein, the defogging module is composed of several Conv modules and DSconv modules, and is used to defog the traffic image data in the foggy environment to generate a clear image.
5. The foggy target detection method based on YOLOv12 optimization according to claim 4, characterized in that: The joint training loss function specifically includes an object detection loss function and an image defogging and restoration loss function. The expression of the joint training loss function is specifically: Loss=α*DetectLoss+β*DehazeLoss Among them, Loss is the joint loss value, DetectLoss is the object detection loss function, DehazeLoss is the image dehazing and restoration loss function, α is the detection weight corresponding to the object detection loss function, and β is the dehazing weight corresponding to the image dehazing and restoration loss function.
6. The method for foggy target detection based on YOLOv12 optimization according to claim 5, characterized in that: The preset improved YOLOv12 model is trained according to the training data to obtain a foggy target detection model, specifically including: Dividing the training data into a training set, a validation set, and a test set by stratified sampling; According to the training set, validation set and test set, the preset improved YOLOv12 model is trained through iterative training until the joint loss value is less than the preset loss threshold, and the training is terminated to obtain the foggy target detection model.
7. The foggy target detection method based on YOLOv12 optimization according to claim 6, characterized in that: The training process also includes: setting a dynamic loss weight strategy; wherein the dynamic loss weight strategy is specifically: During the iterative training process, the detection weight and the defogging weight are set to linearly decrease and linearly increase respectively.
8. The foggy target detection method based on YOLOv12 optimization according to claim 1, characterized in that: The target detection is performed on the target data to be detected by the foggy weather target detection model to obtain the detection result, specifically including: The data to be detected is input into the foggy weather target detection model for defogging and target detection to obtain the detection result; wherein, the detection result includes a clear image corresponding to the data to be detected and the category and prediction frame information of the traffic object contained in the data to be detected, and the prediction frame information includes the center coordinates and the width and height of the traffic object.
9. A foggy target detection device based on YOLOv12 optimization, characterized in that: include: Acquisition module, training module and detection module; The acquisition module is used to acquire raw data and pre-process the raw data to obtain training data; wherein the training data at least includes a plurality of traffic image data in a foggy environment; The training module is used to train a preset improved YOLOv12 model based on the training data to obtain a foggy target detection model; wherein the preset improved YOLOv12 model includes a defogging module and a joint training loss function; The detection module is used to perform target detection on the data to be detected by using the foggy weather target detection model to obtain a detection result.
10. The foggy target detection device based on YOLOv12 optimization according to claim 9, characterized in that: The acquisition module is used to collect raw data and pre-process the raw data to obtain training data, specifically including: Collecting traffic image data under several different environments to obtain the original data; wherein the original data at least includes traffic image data under several foggy environments; The original data is labeled to obtain the training data; wherein the labeling includes the category, center coordinates, object height and width of all traffic objects in the original data.