A mixed traffic flow-based hdv driving intention recognition method and system based on light language

Abstract

The application belongs to the technical field of automatic driving, and particularly relates to a mixed traffic flow-based HDV driving intention recognition method and system based on light language, comprising: constructing an automatic driving network, wherein the automatic driving network comprises a networked automatic driving vehicle CAV, a manually driven vehicle HDV and a MEC node; the CAV acquires a light language image of a surrounding target HDV; the light language image is transmitted to the MEC node, and the MEC node pre-processes the light language image; a trained driving intention recognition model based on a YOLO-v7 network is used to recognize the driving intention of the target HDV from the pre-processed light language image; the HDV driving intention is transmitted to the CAV; the CAV generates a corresponding driving decision according to the HDV driving intention, and executes the driving decision; the YOLO-v7 algorithm is used to recognize the light language signal of the HDV in real time, and the accuracy of the driving intention recognition is improved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a method and system for HDV driving intent recognition based on traffic lights in mixed traffic flow. Background Technology

[0002] As urban traffic environments where connected autonomous vehicles (CAVs) and manually driven vehicles (HDVs) coexist become increasingly common, improving the safety and efficiency of traffic flow has become a key research focus in the transportation field. Intelligent connected vehicles can interact with their surroundings through vehicle-to-everything (V2X) technology. However, in mixed traffic flows, because HDVs cannot communicate directly with CAVs, CAVs struggle to accurately predict and recognize the driving intentions of HDVs, thus affecting their decision-making and driving safety.

[0003] Light signals are the primary means by which HDVs communicate with surrounding vehicles nonverbally, conveying driving intentions through signals such as brake lights and turn signals. However, due to the diversity of light signal patterns and environmental interference, CAVs struggle to accurately identify HDVs' driving intentions using traditional perception methods. Summary of the Invention

[0004] To address the problem of accurate driver intent recognition by autonomous vehicles (CAVs) in mixed traffic flow, this invention proposes a traffic light-based driver intent recognition method for HDVs in mixed traffic flow. The method includes: constructing an autonomous driving network, comprising connected autonomous vehicles (CAVs), manually driven HDVs, and MEC nodes; CAVs acquiring traffic light signals of surrounding target HDVs; transmitting the traffic light signals to the MEC nodes, which preprocess the signals; using a trained YOLO-v7-based driver intent recognition model to identify the target HDV's driver intent from the preprocessed traffic light signals; transmitting the HDV's driver intent to the CAVs; and the CAVs generating and executing corresponding driving decisions based on the HDV's driver intent.

[0005] A traffic signal-based HDV driving intent recognition system for mixed traffic flow includes: a real-time traffic signal image acquisition unit, an HDV driving intent recognition module, a vehicle-to-everything (V2X) intent sharing service module, and a driving behavior decision execution unit.

[0006] The real-time light signal image acquisition unit is used to acquire light signal images from HDV.

[0007] The HDV driving intention recognition module identifies the driving intention of the HDV based on the collected light signal images and uses the driving intention recognition model based on the YOLO-v7 network on the MEC node.

[0008] The vehicle-to-everything (V2X) intent-sharing service module unicasts the HDV's driving intent to the target CAV;

[0009] The driving behavior decision execution unit makes decisions and plans routes based on the received HDV driving intention information, and executes the corresponding driving behaviors.

[0010] The beneficial effects of this invention are:

[0011] This invention employs the YOLO-v7 algorithm for real-time recognition of traffic light signals from HDVs. By introducing an improved feature extraction layer and an optimized network structure, the accuracy of YOLO-v7 in recognizing HDVs' braking, steering, and overtaking intentions in complex traffic environments is enhanced. Utilizing 5G communication technology, traffic light signals are transmitted in real-time from the CAV to the MEC node for processing, and the recognized driving intentions are promptly returned to the CAV for decision-making and execution. Compared to traditional cloud computing methods, using MEC nodes for distributed computing significantly reduces data transmission latency, substantially improves the overall system response speed, and lowers the cost of large-scale deployment. By accurately recognizing HDVs' driving intentions and sharing them with the CAV, the CAV can adjust its driving strategy based on real-time information, avoiding potential traffic accidents and thus improving road traffic safety. Attached Figure Description

[0012] Figure 1 This is a flowchart of the HDV driving intent recognition method of the present invention;

[0013] Figure 2 This is a schematic diagram of the HDV driving intention recognition system of the present invention;

[0014] Figure 3 This is a technical roadmap for the HDV driving intent recognition system of the present invention;

[0015] Figure 4 This is a schematic diagram of the overall HDV driving intention recognition system of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] A method for HDV driver intent recognition based on traffic light signals in mixed traffic flow is disclosed. The method includes: constructing an autonomous driving network, wherein the autonomous driving network includes connected autonomous vehicles (CAVs), manually driven HDVs, and MEC nodes; the CAVs acquire traffic light signal images of surrounding target HDVs; the traffic light signal images are transmitted to the MEC nodes, where the MEC nodes preprocess the traffic light signal images; a trained YOLO-v7-based driver intent recognition model is used to identify the target HDV driver intent from the preprocessed traffic light signal images; the HDV driver intent is transmitted to the CAVs; the CAVs generate corresponding driving decisions based on the HDV driver intents and execute the driving decisions.

[0018] The YOLO-v7 algorithm is a target detection algorithm based on deep convolutional neural networks, widely used in real-time object recognition tasks. YOLO-v7 is renowned for its single-step detection, high accuracy, and speed, significantly improving recognition accuracy while maintaining detection speed. The YOLO-v7 algorithm in this invention further enhances the recognition capability of traffic light signals in complex traffic environments by introducing an improved feature extraction layer and optimized network structure, exhibiting particularly excellent performance in multi-target detection and dynamic backgrounds.

[0019] A specific implementation of a traffic signal-based HDV (High-Depth Vehicle) driver intent recognition method in mixed traffic flow is as follows: Figure 1 and Figure 2 As shown, it specifically includes:

[0020] The S1 and CAV are equipped with high-resolution cameras and other auxiliary sensors, which are positioned at the front and sides of the vehicle to ensure comprehensive coverage of the surrounding environment.

[0021] The camera should have the following characteristics: it should be able to maintain high image quality under complex lighting conditions, such as scenes with strong light and shadow transitions; it should be able to provide a wider field of view to ensure that multiple HDV light signals can be captured simultaneously in multi-lane mixed traffic flow; and it should be able to provide clear images at night or in low light conditions to provide high-quality data for subsequent light signal recognition.

[0022] In this embodiment, when a CAV enters a mixed traffic flow area, the onboard vision sensor begins to capture the light signals of surrounding HDVs in real time. These light signals include brake lights, turn signals, and overtaking signals. The specific acquisition process is as follows: the vision sensor continuously monitors the surrounding environment, identifies and locks onto the source of the HDV light signals, ensuring continuous and stable image acquisition even under high-speed driving and complex traffic scenarios; to ensure that the captured images have sufficient temporal information, the camera's frame rate needs to be automatically adjusted according to actual road conditions to adapt to changes in vehicle speeds and the environment; before being transmitted to the MEC node, the acquired images undergo preliminary processing in the onboard computing unit, including image compression and resolution adjustment, to ensure that the images are not distorted during network transmission.

[0023] S2. The traffic light signal images are transmitted to the MEC node in real time via the 5G communication network, and various image preprocessing operations are performed on the MEC node to improve the recognition accuracy of the traffic light signals. Specifically, the CAV transmits the acquired traffic light signal images to the MEC node in real time via the 5G communication network, ensuring that the traffic light signal image data can be transmitted to the edge computing node quickly and stably in complex traffic environments, reducing transmission latency; combined with error correction technology (Forward Error Correction, FEC) and encrypted transmission, the security and integrity of the image data are guaranteed.

[0024] In this embodiment, after the MEC node receives the light signal image, it performs preprocessing operations, specifically including: improving image contrast through histogram equalization, and using gamma correction to adjust image brightness, making the light signal clearer and more prominent under different lighting conditions; using median filtering to denoise the image, effectively reducing noise interference to light signal recognition while preserving the edge and detail features of the light signal; cropping and scaling the image according to the input requirements of the YOLO-v7 algorithm, adjusting the target light signal area to a suitable size to ensure that the light signal can be correctly and effectively recognized by the YOLO-v7 model; and converting the image from the RGB color space to the HSV or YUV color space. This conversion enhances the color features of the light signal under different lighting conditions, enabling it to be accurately recognized in different environments.

[0025] S3. The YOLO-v7 algorithm is used to analyze the preprocessed traffic light signals to identify various driving intentions of the target HDV, including steering, braking, and overtaking intentions. This process involves multiple steps of image feature extraction, candidate region generation and optimization, target classification and confidence prediction, and finally, the specific driving behavior of the HDV is inferred through comprehensive analysis. The steps are as follows:

[0026] Step 1: Obtain all the light signal images of historical manually driven vehicles to obtain the training set;

[0027] Step 2: Preprocess the light image in the training set by converting the RGB image to YUV or HSV color space and using image enhancement techniques to improve image quality.

[0028] Step 3: Input the preprocessed light signal image into the optimized YOLO-v7 network. Extract multi-scale features from the image through the convolutional neural network to improve the accuracy and adaptability of recognition. Specifically, this includes: enhancing the structure of the convolutional layers to enable the model to extract spatial and contextual features of the light signal from different scales; and increasing the depth of the convolutional layers and improving feature extraction efficiency to ensure accurate capture of changes in the light signal in complex traffic environments.

[0029] Step 4: Use the Region Proposal Network (RPN) to process multi-scale features and introduce adaptive candidate box generation technology to dynamically generate more suitable candidate box sizes based on actual traffic conditions and historical data, thus generating candidate regions.

[0030] Step 5: Classify each candidate region and identify whether there is a light signal in the candidate region. If there is, proceed to step 6; otherwise, discard the candidate region.

[0031] Step 6: Identify the driving status of manually driven vehicles based on light signals, classify the light signals, and identify the driving intentions;

[0032] Step 7: A weighted strategy is introduced into the loss function of the YOLO-v7 model, that is, the classification loss, bounding box localization loss and confidence loss are optimized by weighting and dynamically adjusting their weights; the loss function of the model is calculated, and an L2 regularization term is introduced to prevent the model from overfitting during training; the Adam optimization algorithm is used to optimize the model parameters, and the model training is completed when the loss function converges.

[0033] Specifically, this involves: First, inputting the preprocessed light image into the YOLO-v7 model. The preprocessed image is first processed through a deep convolutional neural network in YOLO-v7 for feature extraction. This step extracts the spatial and semantic features of the image step by step through multiple convolutional layers.

[0034] G i =φ(W i ·X+c i )

[0035] Among them, G i W represents the output feature map of the i-th layer; i and c i φ represents the weights and biases of the i-th layer, respectively; X is the input image or the output of the previous layer; φ is the activation function (ReLU).

[0036] Each convolutional layer extracts features from low to high levels, forming multi-scale feature maps. After a series of convolutional operations, these feature maps become more refined and abstract, and will be used for subsequent region proposal and object detection.

[0037] G i+1 =φ(W i+1 ·G i +c i+1 )

[0038] As network depth increases, the spatial resolution of the feature maps decreases, but the semantic information increases. After feature extraction, the resulting feature maps contain rich spatial and semantic information from the image.

[0039] After feature map extraction, the YOLO-v7 RPN network generates multiple anchor boxes based on the feature maps from the previous layer. These anchor boxes represent the locations in the image where target objects may exist. Each anchor box undergoes classification and regression calculations to determine whether it contains a target.

[0040]

[0041]

[0042]

[0043]

[0044] Where Δx is the horizontal coordinate offset of the predicted bounding box; x is the horizontal coordinate of the center of the predicted bounding box of the target object; x anchor w represents the horizontal coordinate of the anchor frame center. anchor The width of the anchor box; Δy is the vertical coordinate offset of the prediction box; y is the vertical coordinate of the center of the prediction box of the target object; y anchor h is the vertical coordinate of the anchor frame center; anchor Δw is the height of the anchor frame; Δw is the normalized logarithmic offset of the prediction box width relative to the anchor frame; w pred h is the width of the predicted bounding box of the target object; Δh is the normalized logarithmic offset of the predicted bounding box height relative to the anchor box; pred h is the height of the predicted bounding box of the target object. anchor This represents the height of the anchor frame.

[0045] The generated anchor boxes will be further optimized through regression analysis to more accurately fit the location of the target light signal. This optimization is achieved through a regression loss function:

[0046]

[0047] Among them, L regThe regression loss function measures the difference in position and size between the predicted bounding box and the ground truth bounding box; N pos Δx represents the number of positive samples, i.e., the number of samples correctly identified as the target. i Δy i Δw i ,Δh i The regression offset of the prediction box represents the center coordinate offset and the adjustment of the width and height of the prediction box, respectively. These are the regression parameters for the actual target bounding boxes, used to compare them with the predicted values. The model uses these parameters multiple times in each iteration during training until the final loss function value converges or becomes sufficiently small, indicating that the predicted box location is accurate enough.

[0048] Based on the optimized anchor box, YOLO-v7 further refines the location of the light signal using bounding box regression. The regression formula is as follows:

[0049] BBox(x,y,w,h)=(φ(Δx)+x anchor ,φ(Δy)+y anchor ,e Δw ·w anchor ,e Δh ·h anchor )

[0050] Where BBox(x,y,w,h) is the final predicted bounding box, represented by the center coordinates (x,y) of the bounding box, as well as its width w and height h; φ(Δx)+x anchor and φ(Δy)+y anchor Here are the final horizontal and vertical coordinates of the center point of the predicted bounding box; φ is the activation function (ReLU); x anchor and y anchor These are the center coordinates of the anchor frame; e Δw ·w anchor and e Δh ·h anchor To predict the width and height of the bounding box, multiply it by the width w of the anchor box after exponential scaling. anchor and height h anchor This yields the final dimensions. The bounding box not only represents the location of the signal light but also includes its size information. The anchor box position and size are adjusted to more closely approximate the actual target's location and size.

[0051] A regression method based on mean squared error (MSE) is used to optimize the center coordinates and size errors of the bounding box. The expression for the bounding box localization loss is as follows:

[0052]

[0053] Among them, Lcoord The mean squared error (MSE) for bounding box localization is used to evaluate the difference between the predicted and actual bounding boxes; S 2 B represents the number of grids into which the image is segmented; B represents the number of bounding boxes predicted for each grid. The function is an indicator function; it takes a value of 1 when the j-th anchor box in the i-th cell contains the target object, and a value of 0 otherwise. i y i These are the center coordinates of the actual bounding box; To predict the center coordinates of the bounding box; w i h i This represents the width and height of the actual bounding box; To predict the width and height of the bounding box.

[0054] For each generated bounding box, YOLO-v7 classifies the target using the Softmax function to determine which signal it belongs to (e.g., turn, brake, overtaking, etc.). The classification formula is:

[0055]

[0056] Where, η i σ(η) represents the predicted score for category i; k represents the total number of categories; σ(η) represents the predicted score for category i. i ) represents the probability of the i-th category, calculated using the Softmax function.

[0057] Classification loss L cls This is used to measure the model's classification accuracy in the driving intent recognition task. Specifically, this loss function evaluates the model's classification ability by calculating the weighted negative log-likelihood between the predicted label and the true label. That is, it introduces a class weight α. c This allows different categories of light signals to be assigned different weights based on their frequency or importance, thus solving the problem of data imbalance, as expressed below:

[0058]

[0059] Where, α c These are class weights, used to handle class imbalance problems; y ic This represents the true label (value 0 or 1) indicating that the i-th sample belongs to the c-th class; The probability that the i-th sample is predicted to be of class c after adjustment by the activation function; N represents the total number of samples; and c represents the total number of classes. By introducing class weights and activation functions to calculate the classification loss, class imbalance is handled during model training, and the model's classification accuracy for various driving intentions (such as steering, braking, overtaking, etc.) is optimized.

[0060] In addition to classification, YOLO-v7 also needs to predict the confidence level of each bounding box containing a target object. The confidence level determines the model's certainty about the presence of the light signal. A penalty coefficient λ is introduced for the case where no object is present. noobj To reduce false detections of invalid targets by the network, the confidence loss is calculated using the following formula:

[0061]

[0062] Among them, L conf C is the confidence loss function, used to evaluate the difference between the confidence level that the predicted bounding box contains the target object and the actual situation; i The confidence score is the predicted value, representing the probability that a target object exists within the bounding box. This represents the actual confidence level (ideally 1 or 0). This is an indicator function that takes a value of 1 when the anchor box contains the target object; This is an indicator function that takes a value of 1 when the anchor frame does not contain the target object; λ noobj This is the penalty coefficient in the absence of objects, used to reduce the impact of confidence errors on irrelevant regions.

[0063] After completing classification and confidence prediction, YOLO-v7 combines the location, category, and confidence information of the light signals to infer the HDV's driving intention. By integrating the classification results of multiple candidate boxes, the system infers the HDV's current driving behavior, such as braking, steering, or overtaking. The formula for intention analysis is as follows:

[0064]

[0065] Where I represents the overall driving intention score; N is the total number of candidate boxes; p i It is the predicted probability of the i-th candidate box; These are indicator functions, indicating whether the candidate box represents a turn signal, braking signal, or overtaking signal; L turn L brake L overtake These represent the feature confidence levels of steering, braking, and overtaking signals, respectively.

[0066] The YOLO-v7 algorithm combines classification loss, confidence loss, and bounding box localization loss into a single loss function, which is then optimized through backpropagation. This process ensures the accuracy and reliability of the model across various driving intent recognition tasks.

[0067] To enhance the model's generalization ability, an L2-norm-based regularization term is added to the loss function to control model complexity. The final optimized loss function is:

[0068] L=λcoord L coord +λ conf L conf +λ cls L cls +λ reg ·W 2

[0069] Where, λ coord The weighting coefficients for the bounding box localization loss; L coord The bounding box localization loss measures the error in the position and size of the bounding box; λ conf L is the weighting coefficient for confidence loss. conf λ represents the confidence loss, which measures the probability error of the target's existence. cls L represents the weighting coefficients for the classification loss. cls The classification loss measures the accuracy of the target classification; λ reg It is the weight of the regularization term; W 2 The L2 norm, representing the model parameters, is used to reduce the risk of overfitting. Finally, the Adam optimization algorithm is used to continuously adjust the model parameters, minimizing the loss function and thus improving the model's recognition performance.

[0070] S4. Through the aforementioned steps, the HDV's driving intentions (steering, braking, overtaking) have been identified using the YOLO-v7 algorithm. To ensure the CAV can react promptly, these identified driving intentions are transmitted in real-time to the CAV's OBU terminal via a 5G communication network. Utilizing the high bandwidth and low latency characteristics of the 5G network, the identified driving intentions are transmitted to the target CAV via unicast, ensuring the timeliness and reliability of data transmission.

[0071] In this embodiment, after receiving the HDV's driving intention information, the CAV's OBU terminal first parses the data. The CAV combines the received driving intention with its own sensor data and uses its built-in decision support system to analyze the current environment. The decision support system considers multiple factors: the HDV's intention, the traffic conditions in the current lane, the traffic flow in adjacent lanes, and the CAV's own driving status. Based on these analysis results, the CAV generates multiple feasible driving strategies and selects the optimal strategy through a path planning algorithm. After selecting the optimal driving strategy, the CAV's control system generates specific driving behavior instructions based on that strategy. These instructions include deceleration, lane changing, acceleration, or emergency braking. The control system adjusts the vehicle's powertrain, steering, and braking systems accordingly to ensure the CAV drives safely according to the optimal decision. During the execution of driving behavior, the CAV's sensor system continuously monitors the surrounding environment and its own status. If the environment changes, such as a vehicle ahead suddenly decelerating or changing lanes, the CAV immediately reassesses the current situation and adjusts its driving strategy through the decision support system to ensure safe driving. Meanwhile, the execution results will be transmitted to the CAV's control system and vehicle networking system through a feedback mechanism to optimize future decision-making processes.

[0072] In this embodiment, a feedback mechanism is established to formulate corresponding driving strategies. The intelligent connected vehicle (OBU) sends feedback to the MEC node based on the received driving intention information and actual driving conditions, which is used for continuous learning and optimization of the model.

[0073] A traffic signal-based HDV driving intent recognition system for mixed traffic flow includes: a real-time traffic signal image acquisition unit, an HDV driving intent calculation module, a vehicle-to-everything (V2X) intent sharing service module, and a driving behavior decision execution unit. These units work together to achieve real-time recognition and response to HDV driving intent in mixed traffic flow.

[0074] The real-time traffic light signal image acquisition unit consists of onboard vision sensors installed on CAVs (Connected Autonomous Vehicles). This unit uses high-definition cameras to acquire real-time traffic light signal images from surrounding HDVs (Human-Driven Vehicles). These vision sensors can accurately capture the traffic light signal information of HDVs under different lighting conditions (such as daytime, nighttime, and rainy / foggy weather) and transmit this image data to the onboard computing unit for preliminary processing in real time. The acquired traffic light signal image data is transmitted to MEC (Edge Computing) nodes via a 5G communication network to ensure fast and stable data transmission in complex traffic environments.

[0075] The HDV driving intent calculation module is deployed in the MEC node. This module first preprocesses the received light signal image data, including image enhancement, denoising, cropping, and color space conversion, to improve image quality and the accuracy of light signal recognition. Then, it uses the YOLO-v7 algorithm to extract and analyze features from the preprocessed image, identifying the HDV's specific driving intent, such as steering, braking, and overtaking intentions. Through this module's processing, the system can efficiently and accurately identify HDV driving behavior, providing reliable data support for CAV decision-making.

[0076] The vehicle-to-everything (V2X) intent-sharing service module unicasts the identified HDV driving intent information to the target CAV's OBU (On-Board Unit) terminal via a 5G communication network. This module leverages the high bandwidth and low latency of the 5G network to ensure that driving intent information is transmitted to the target CAV in the shortest possible time. Through this module, the CAV can receive driving intents from surrounding HDVs in real time and make corresponding driving decisions based on the received information, ensuring safety and efficiency in complex traffic environments.

[0077] The driving behavior decision execution unit is located inside the CAV. This unit receives HDV driving intention information from the OBU and combines it with the vehicle's own sensor data and current traffic conditions, using a decision support system for comprehensive analysis. Based on the analysis results, the unit generates optimal driving behavior strategies, such as deceleration, acceleration, lane changing, or emergency braking. These driving commands are executed by the CAV's control system to ensure safe driving according to the optimal strategy. Furthermore, this unit has a dynamic adjustment function, capable of adjusting driving behavior according to real-time environmental changes to cope with emergencies, further enhancing the CAV's ability to handle mixed traffic flow.

[0078] In one specific embodiment of the present invention, such as Figure 3As shown, the CAV (Consumer-Aided Vehicle) uses onboard vision sensors to collect real-time images of traffic light signals from surrounding HDVs (Hyper-Driven Vehicles). Once the target HDV's traffic light signal is successfully detected, the system transmits the collected image data to the MEC (Edge Computing) node via a 5G communication network. At the MEC node, the image data undergoes preprocessing, including noise reduction and image enhancement, to improve the quality of the traffic light signals and the accuracy of subsequent recognition. The preprocessed image data is then used for driving intention recognition via the YOLO-v7 algorithm. This algorithm can identify the HDV's specific driving intentions, including steering, braking, and overtaking intentions. The identified driving intentions are unicast to the target CAV's OBU (On-Board Unit) via the 5G network. Upon receiving the intention, the OBU, combined with current environmental information, makes corresponding driving decisions, including lane changing, deceleration, or overtaking. Finally, the CAV executes the corresponding driving operation and shares driving information with surrounding vehicles through the cooperative driving system, thereby improving traffic safety and efficiency. The entire system processes information efficiently at the MEC node, enabling rapid response and adjustment of the CAV's driving behavior, achieving intelligent perception and decision-making in complex traffic environments. Overall schematic diagram as follows Figure 4 As shown.

[0079] The system in this embodiment of the invention can effectively identify and share the driving intentions of HDVs in mixed traffic flow, and improve the overall safety and efficiency of traffic through the intelligent decision-making and execution of CAVs.

[0080] Table 1 YOLO-v7 Parameter Settings

[0081]

[0082]

[0083] Table 2 Training and Test Dataset Settings

[0084]

[0085] Based on the model parameters and training / test dataset settings, the recognition results are shown in Table 3:

[0086] Table 3 Model Evaluation Indicators

[0087]

[0088] Table 3 shows that the model achieves an average accuracy of 68.5% at a 0.5 IoU threshold, and an average accuracy of 45.7% across multiple more stringent IoU threshold ranges. These results demonstrate the model's high accuracy in target localization, particularly at lower IoU thresholds, where it achieves an accuracy of 91.3% and a recall of 89.5%. This indicates that in real-world traffic scenarios, the model can accurately and effectively identify most driving intentions. Furthermore, the 15ms / frame inference time ensures real-time system response, which is significant for the practical application of intelligent connected vehicles. This patent embodiment, through optimization and parameter tuning of the YOLO-v7 model, successfully implements a highly efficient driving intention recognition system. The model exhibits strong recognition capabilities in complex traffic flow environments, providing timely and accurate driving intention information for intelligent connected vehicles, supporting safe and reasonable driving decisions. These evaluation results not only meet practical application requirements but also demonstrate the practical application value of this invention in intelligent transportation systems.

[0089] The system implementation method of the present invention is the same as the method implementation method.

[0090] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include ROM, RAM, disk, or optical disk, etc.

[0091] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for HDV driver intent recognition based on traffic lights in mixed traffic flow, characterized in that, include: An autonomous driving network is constructed, comprising connected autonomous vehicles (CAVs), manually driven vehicles (HDVs), and MEC nodes. The CAVs acquire traffic light signals from surrounding HDVs; these signals are transmitted to the MEC nodes, which preprocess them. A trained YOLO-v7-based driving intent recognition model is used to identify the driving intent of the target HDVs from the preprocessed traffic light signals; finally, the HDV driving intent is transmitted back to the CAVs. CAV generates corresponding driving decisions based on HDV's driving intentions and executes those decisions. Training the driving intention recognition model based on the YOLO-v7 network includes: Step 1: Obtain all the light signal images of historical manually driven vehicles to obtain the training set; Step 2: Preprocess the light language images in the training set, that is, convert the RGB images to YUV or HSV color space, and use image enhancement technology to enhance the images after the color space conversion. Step 3: Input the preprocessed light image into the optimized YOLO-v7 network and extract multi-scale features from the image through a convolutional neural network; Step 4: Use the Region Proposal Network (RPN) to process multi-scale features, that is, introduce an adaptive candidate box generation technique to dynamically generate candidate boxes of corresponding sizes based on actual traffic conditions and historical data; and select candidate regions based on the candidate boxes. Specifically, this includes: generating multiple anchor boxes of different sizes and aspect ratios for each pixel in the feature map; for each anchor box, predicting the probability and class of containing the target, and simultaneously predicting the precise bounding box coordinates of the target; the formula is as follows: Where Δx is the horizontal coordinate offset of the predicted bounding box; x is the horizontal coordinate of the center of the predicted bounding box of the target object; x anchor w represents the horizontal coordinate of the anchor frame center. anchor Δy is the width of the anchor box; Δy is the vertical coordinate offset of the prediction box; y is the vertical coordinate of the center of the prediction box of the target object; y anchor h is the vertical coordinate of the anchor frame center; anchor Δw is the height of the anchor frame; Δw is the normalized logarithmic offset of the prediction box width relative to the anchor frame; w pred h is the width of the predicted bounding box of the target object; Δh is the normalized logarithmic offset of the predicted bounding box height relative to the anchor box; pred h is the height of the predicted bounding box of the target object. anchor The height of the anchor frame; Step 5: Classify each candidate region and identify whether there is a light signal in the candidate region. If there is, proceed to step 6; otherwise, discard the candidate region. Step 6: Classify the light signals according to the driving status of the manually driven vehicle and identify its driving intention; Step 7: Calculate the loss function of the YOLO-v7 model based on the driving intention. The loss function incorporates a weighted strategy, which optimizes the classification loss, bounding box localization loss, and confidence loss by dynamically adjusting their weights. An L2 regularization term is also included in the loss function. The Adam optimization algorithm is used to optimize the model parameters. When the loss function converges, the model training is complete.

2. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, The target HDV's light signal image includes brake lights, turn signals, and overtaking signal lights.

3. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, The light signal images are transmitted to the MEC node in real time via 5G communication.

4. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, The preprocessing of the light message image includes: enhancing the light message image using histogram equalization and gamma correction techniques; performing median filtering on the enhanced light message image; cropping and scaling the filtered image; and converting the RGB color space of the cropped and scaled image to HSV or YUV color space to obtain the preprocessed light message image.

5. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, Convolutional neural networks extract multi-scale features from images as follows: G i =φ(W i ·X+c i ) G i+1 =φ(W i+1 ·G i +c i+1 ) Among them, G i W represents the output feature map of the i-th layer; i and c i These are the weights and biases of the i-th layer, respectively; X is the input image or the output of the previous layer; φ is the activation function.

6. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, Identifying the driving status of a manually driven vehicle involves calculating the confidence level of the driver's intention and obtaining the driver's intention based on the confidence level value. Its calculation expression is: Where I represents the overall driving intention score; N is the total number of candidate boxes; p i It is the predicted probability of the i-th candidate box; These are indicator functions, indicating whether the candidate box represents a turn signal, braking signal, or overtaking signal; L turn L brake L overtake These represent the feature confidence levels of steering, braking, and overtaking signals, respectively.

7. The HDV driving intent recognition method based on traffic lights in mixed traffic flow according to claim 1, characterized in that, The model's loss function includes bounding box localization loss, confidence loss, and classification loss; A regression method based on mean squared error is used to optimize the center coordinates and size errors of the bounding box; that is, the expression for the bounding box localization loss is: Among them, L coord Mean square error for bounding box localization; S 2 B represents the number of grids into which the image is segmented; B represents the number of bounding boxes predicted for each grid. For indicator functions; x i y i These are the center coordinates of the actual bounding box; To predict the center coordinates of the bounding box; w i h i This represents the width and height of the actual bounding box; To predict the width and height of the bounding box; Introducing the penalty coefficient λ in the case of no objects noobj To reduce false detections of invalid targets by the network, the confidence loss expression is as follows: Among them, L conf C is the confidence loss function; i The confidence score for the prediction; This represents the actual confidence level. Both are indicator functions; λ noobj This represents the penalty coefficient in the case of no objects. Introducing category weight α c Different weights are assigned based on the frequency or importance of different categories of light signals, where the classification loss expression is: Where, α c It is the category weight; y nc This represents the true label of the nth sample belonging to the cth class; It is the probability that the nth sample is predicted to be of class c after adjustment by the activation function; N represents the total number of samples; C represents the total number of classes; The optimized model loss function is: L=λ coord L coord +λ conf L conf +λ cls L cls +λ reg ·||W|| 2 Where, λ coord The weighting coefficients for the bounding box localization loss; L coord For bounding box localization loss; λ conf L is the weighting coefficient for confidence loss. conf For confidence loss; λ cls L represents the weighting coefficients for the classification loss. cls For classification loss; λ reg It is the weight of the regularization term; ||W|| 2 The L2 norm represents the model parameters.

8. A traffic signal-based HDV driving intention recognition system for mixed traffic flow, the system being used to execute the traffic signal-based HDV driving intention recognition method for mixed traffic flow as described in any one of claims 1 to 7, characterized in that, include: Real-time light signal image acquisition unit, HDV driving intention recognition module, vehicle network intention sharing service module, and driving behavior decision execution unit; The real-time light signal image acquisition unit is used to acquire light signal images from HDV. The HDV driving intention recognition module identifies the driving intention of the HDV based on the collected light signal images and uses the driving intention recognition model based on the YOLO-v7 network on the MEC node. The vehicle-to-everything (V2X) intent-sharing service module unicasts the HDV's driving intent to the target CAV; The driving behavior decision execution unit makes decisions and plans routes based on the received HDV driving intention information, and executes the corresponding driving behaviors.

Images