A vehicle platoon visual interactive adjustment method and device and a storage medium

Abstract

The application relates to a vehicle platoon visual interactive adjustment method and device and a storage medium, and belongs to the intelligent transportation field, wherein the method comprises the following steps: receiving, from a lead vehicle, a platoon posture expected visual point cloud of a first vehicle and a first follower vehicle; collecting first follower vehicle image information through a rear binocular image collector of the first vehicle and collecting first vehicle image information through a front binocular image collector of the first follower vehicle; performing platoon posture recognition based on the collected image information to obtain a platoon posture monitoring visual point cloud; and performing rendering identification on the monitoring point cloud based on difference information of the expected point cloud and the monitoring point cloud, and sending the monitoring point cloud to the lead vehicle through a V2V communication network to perform platoon optimization. Through binocular vision point cloud processing based on an edge node and a non-periodic V2V communication mechanism driven by differences, the technical effects of saving communication resources, avoiding network congestion and realizing accurate dynamic platoon posture optimization adjustment based on visual point cloud differences are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation, and in particular to a vehicle platoon visual interactive adjustment method and device and a storage medium. BACKGROUND

[0002] With the rapid development of intelligent transportation systems, vehicle platoon control technology has become an important means to improve road traffic efficiency and safety. Vehicle platoon can effectively reduce traffic congestion, reduce energy consumption and improve road utilization by cooperative driving of multiple vehicles.

[0003] Currently, existing vehicle platoon control technology mainly adopts a cooperative control method based on a vehicle-mounted communication network. In traditional platoon control systems, each vehicle node usually adopts a periodic communication mechanism with a fixed frequency to exchange state information such as position, speed, and acceleration, in order to achieve coordinated control of the platoon. The lead vehicle is responsible for planning the platoon path and speed, and the following vehicles adjust their driving state according to the received information to maintain the preset platoon formation. However, the traditional periodic communication mechanism has the problem of waste of communication resources. Whether the state of the vehicle platoon changes or not, information exchange will be carried out at a fixed time interval, which leads to a large amount of communication data being redundant in the case of relatively stable vehicle state, causing inefficient occupation of network bandwidth. Moreover, frequent periodic communication can easily cause network congestion and communication delay, especially in multi-vehicle platoon or complex traffic environment, intensive data exchange can cause excessive load on the communication network, affecting the real-time transmission of critical control information, and thus affecting the stability and safety of platoon control. In addition, existing platoon control is mostly based on simple position and speed information for adjustment, lacking accurate visual monitoring and analysis capability of the vehicle platoon posture, and it is difficult to achieve accurate adjustment based on the difference between the actual platoon state and the expected state, affecting the precision and adaptability of platoon control. SUMMARY

[0004] The present application provides a vehicle platoon visual interactive adjustment method, device and storage medium to solve the technical problems in the prior art that the fixed frequency periodic communication of vehicle platoon control leads to inefficient occupation of network bandwidth, communication delay congestion, and inability to achieve accurate visual posture adjustment.

[0005] The technical solution of the present application to solve the above technical problems is as follows:

[0006] In a first aspect, the application provides a vehicle platoon visual interactive adjustment method applied to a first edge node embedded in a first follower vehicle, comprising: receiving, from a lead vehicle, a platoon posture expected visual point cloud of a first vehicle and a first follower vehicle, wherein the first follower vehicle is a follower vehicle of the first vehicle, and the first vehicle is the lead vehicle or a follower vehicle; collecting first follower vehicle image information from a rear binocular image collector of the first vehicle, and collecting first vehicle image information from a front binocular image collector of the first follower vehicle; performing platoon posture recognition based on the first follower vehicle image information and the first vehicle image information to obtain a platoon posture monitoring visual point cloud; rendering and marking the platoon posture monitoring visual point cloud based on difference point cloud information of the platoon posture expected visual point cloud and the platoon posture monitoring visual point cloud, and sending the platoon posture monitoring visual point cloud to the lead vehicle through a V2V communication network to perform vehicle platoon optimization.

[0007] In a second aspect, the application provides an electronic device, comprising: a memory for storing a computer software program; and a processor for reading and executing the computer software program, thereby implementing the vehicle platoon visual interactive adjustment method provided by the application.

[0008] In a third aspect, the application provides a computer-readable storage medium having a computer software program stored therein, wherein the computer software program is executed by a processor to implement the vehicle platoon visual interactive adjustment method provided by the application.

[0009] The application has the following beneficial effects:

[0010] The application is applied to a first edge node embedded in a first follower vehicle, and by sinking the computing processing capability to the vehicle edge node, distributed platoon control processing is realized, thereby reducing the computing burden of the center node and improving the response speed. The platoon posture expected visual point cloud of the first vehicle and the first follower vehicle is received from a lead vehicle, wherein the first follower vehicle is a follower vehicle of the first vehicle, and the first vehicle is the lead vehicle or a follower vehicle. By establishing an expected reference of platoon control, the lead vehicle uniformly plans and distributes the expected state of the platoon posture, so as to ensure that each vehicle node has a consistent control target. The first follower vehicle image information is collected from a rear binocular image collector of the first vehicle, and the first vehicle image information is collected from a front binocular image collector of the first follower vehicle. The real-time visual information of adjacent vehicles in the platoon is obtained through bidirectional image collection, and the binocular image collector can provide depth information, thereby providing a data basis for subsequent three-dimensional posture recognition.

[0011] The formation posture recognition is performed based on the first follower vehicle image information and the first vehicle image information, the formation posture monitoring visual point cloud is obtained, the two-dimensional image information is converted into three-dimensional point cloud data, the accurate quantitative representation of the current formation posture is realized, and the computable data basis is provided for posture comparison and adjustment. The difference point cloud information between the formation posture expected visual point cloud and the formation posture monitoring visual point cloud is used to render and identify the formation posture monitoring visual point cloud and send it to the leading vehicle through the V2V communication network, so that the vehicle formation optimization is performed, the formation deviation is identified through the point cloud difference analysis, the communication transmission is only performed when there is a significant difference, the on-demand communication of the non-periodic type is realized, and the formation adjustment is more intuitive and accurate through the visual rendering and identification.

[0012] Through the technical scheme, the distributed formation control based on edge computing is realized, the binocular vision and point cloud processing technology are used for accurate posture monitoring, and the non-periodic communication mechanism driven by the difference is used, so that the resource waste problem of the traditional periodic communication is effectively solved, and the visual accurate formation adjustment capability is provided. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 A flowchart of a vehicle formation visual interactive adjustment method provided by the application is shown in the figure.

[0014] Figure 2 A structure diagram of an electronic device provided by the application is shown in the figure.

[0015] Figure 3 A structure diagram of a computer readable storage medium provided by the application is shown in the figure.

[0016] In the drawings, the components represented by the numbers are listed as follows:

[0017] The electronic device 100, the memory 110, the processor 120, the computer software program 111, and the computer readable storage medium 200. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] Example 1, as Figure 1 As shown, this application embodiment provides a vehicle platooning visualization and interactive adjustment method, which is applied to a first edge node embedded in a first following vehicle.

[0022] Specifically, this application embodiment is applied to a first edge node embedded in a first following vehicle. The first edge node refers to a computing device integrated on the first following vehicle, a hardware unit with data processing, image analysis, and communication functions. This first edge node adopts an embedded architecture, including a processor, memory, communication module, and a dedicated image processing acceleration unit, enabling it to perform complex computing tasks locally within the vehicle, reducing dependence on a central server and improving response speed.

[0023] The first following vehicle refers to an intelligent vehicle that follows the vehicle in front in a platoon. It is equipped with perception, decision-making, and execution systems and can autonomously adjust its driving status to maintain the platoon formation. Through an embedded first edge node, the first following vehicle can not only receive instructions from the lead vehicle, but also make intelligent decisions based on local perception information, thereby achieving distributed platoon control.

[0024] The methods for visually adjusting vehicle formations include:

[0025] S1. Receive the desired visualization point cloud of the formation attitude of the first vehicle and the first following vehicle from the lead vehicle, wherein the first following vehicle is the vehicle following the first vehicle, and the first vehicle is the lead vehicle or the following vehicle.

[0026] Specifically, the first edge node receives the desired formation attitude visualization point cloud from the lead vehicle. This desired formation attitude visualization point cloud is a three-dimensional point cloud data pre-calculated and generated by the lead vehicle, used to characterize the spatial positional relationship and attitude configuration between the first vehicle and the first following vehicle in an ideal formation state.

[0027] In this formation, the first following vehicle is the vehicle that follows the first vehicle and is located after the first vehicle in the formation sequence. The first vehicle in the formation can be either a lead vehicle or a follower vehicle. When the first vehicle is a lead vehicle, the first following vehicle is the second vehicle in the formation; when the first vehicle is a follower vehicle, the first following vehicle is the vehicle located further back in the formation.

[0028] The lead vehicle transmits the desired formation attitude visualization point cloud to the first edge node embedded in the first following vehicle via a V2V communication network. This desired formation attitude visualization point cloud contains key parameter information such as the relative position coordinates, vehicle attitude angles, and inter-vehicle distances of the first vehicle and the first following vehicle in the ideal formation state, providing a desired target benchmark for subsequent formation attitude monitoring and adjustment control.

[0029] S2. Collect image information of the first following vehicle from the binocular image acquisition device behind the first vehicle, and collect image information of the first vehicle from the binocular image acquisition device in front of the first following vehicle.

[0030] Specifically, a binocular image acquisition device installed behind the first vehicle acquires image information of the first following vehicle in real time; at the same time, a binocular image acquisition device installed in front of the first following vehicle acquires image information of the first vehicle in real time.

[0031] The binocular image acquisition unit behind the first vehicle is located in the rear area of ​​the first vehicle, with its optical lens facing the rear of the vehicle, and is used to capture the appearance image of the first following vehicle. The binocular image acquisition unit in front of the first following vehicle is installed at the front of the first following vehicle, with its optical lens facing the front of the vehicle, and is used to capture the appearance image of the first vehicle in front.

[0032] The binocular image acquisition device uses stereo vision technology, simultaneously acquiring images through two cameras (left and right) to obtain three-dimensional spatial information of the target vehicle. The image information of the first following vehicle includes visual features such as the front outline, body posture, and relative position of the first following vehicle; the image information of the first vehicle includes visual features such as the rear outline, body posture, and relative position of the first vehicle.

[0033] By acquiring images from both the front and rear, real-time visual information of two adjacent vehicles in the formation can be obtained simultaneously, providing the necessary image data foundation for subsequent formation attitude recognition and relative position calculation.

[0034] S3. Based on the image information of the first following vehicle and the image information of the first vehicle, perform formation posture recognition to obtain a visualized point cloud for formation posture monitoring.

[0035] Specifically, the first edge node uses image recognition algorithms to identify the formation posture based on the collected image information of the first following vehicle and the first vehicle image information, and finally generates a visualized point cloud for formation posture monitoring.

[0036] Formation attitude recognition refers to the analysis of acquired binocular images using computer vision technology to identify and extract formation parameters such as the actual spatial position relationship, vehicle attitude angles, and relative distances between the first vehicle and the first following vehicle. Utilizing the stereoscopic vision characteristics of binocular images, combined with vehicle appearance feature recognition algorithms, real-time formation attitude information between the two vehicles is calculated. The formation attitude monitoring visualization point cloud is the output of formation attitude recognition. This visualization point cloud represents the actual spatial configuration of the first vehicle and the first following vehicle in the form of three-dimensional coordinate points, corresponding to the expected visualization point cloud of formation attitude. The formation attitude monitoring visualization point cloud reflects the current true formation attitude obtained through real-time image recognition, including parameters such as the actual vehicle position coordinates, actual attitude angles, and actual vehicle-to-vehicle distances.

[0037] By converting image information into a standardized point cloud format, a unified data representation can be provided for subsequent formation attitude comparison analysis and optimization adjustments.

[0038] S4. Based on the difference point cloud information between the desired formation attitude visualization point cloud and the formation attitude monitoring visualization point cloud, render and mark the formation attitude monitoring visualization point cloud and send it to the lead vehicle through the V2V communication network to perform vehicle formation optimization.

[0039] Specifically, the first edge node compares and analyzes the visual point cloud of the formation attitude expectation with the visual point cloud of the formation attitude monitoring, calculates the difference point cloud information, and performs rendering and labeling processing on the visual point cloud of the formation attitude monitoring based on the difference information. Finally, the processing result is sent to the lead vehicle through the V2V communication network to perform formation optimization control.

[0040] The difference point cloud information refers to the deviation data between the desired formation attitude visualization point cloud and the formation attitude monitoring visualization point cloud. This deviation reflects the degree of deviation of key parameters such as spatial position differences, attitude angle differences, and inter-vehicle distance differences between the ideal formation state and the actual formation state. Through point cloud registration algorithms and difference calculation methods, the corresponding point position differences between the two sets of point cloud data are quantitatively analyzed to generate difference point cloud information containing deviation vectors and deviation amplitudes. Subsequently, based on the difference point cloud information, the formation attitude monitoring visualization point cloud is rendered and labeled, that is, deviation identification information, such as deviation direction indicators, deviation degree markers, and key adjustment area annotations, is embedded into the original monitoring point cloud data, facilitating the lead vehicle to quickly identify and process formation adjustment needs.

[0041] The rendered point cloud for platoon attitude monitoring is sent to the lead vehicle via a V2V communication network. After receiving the point cloud data with deviation markers, the lead vehicle performs vehicle platoon optimization in conjunction with the global platoon control strategy. This includes generating and issuing control commands such as adjusting target vehicle speed, correcting driving trajectory, and optimizing inter-vehicle distance, thereby achieving dynamic adjustment and optimization control of platoon attitude.

[0042] The above technical solution triggers the communication and adjustment process only when a formation attitude deviation is detected, avoiding redundant data transmission of traditional periodic communication mechanisms. This achieves the technical effects of saving communication resources, avoiding network congestion, and realizing precise dynamic formation attitude optimization and adjustment based on visualized point cloud differences.

[0043] Furthermore, the desired formation attitude visualization point cloud received from the lead vehicle for the first vehicle and the first following vehicle includes:

[0044] S11. Using a navigation vehicle, download the target road features from a cloud map, wherein the target road features include road slope, road type, and road surface point cloud model;

[0045] S12. Obtain the vehicle types and vehicle loads in the formation;

[0046] S13. Retrieve the high-frequency speed and high-frequency distance of vehicle samples that satisfy the road slope, road type, platoon vehicle model and platoon vehicle load.

[0047] S14. Based on the high-frequency speed and high-frequency distance of the vehicle samples, and combined with the number of vehicles in the formation, retrieve the vehicle point cloud model, deploy the vehicle formation from the starting point to the ending point on the road point cloud model, and obtain the visual point cloud of the formation attitude expectation.

[0048] In one feasible implementation, the first edge node establishes a connection with the cloud map server through the onboard communication module of the lead vehicle to download the target road features corresponding to the current driving path. These target road features include road slope, road type, and road surface point cloud model. Road slope represents the longitudinal slope variation parameter of the road; road type information includes road condition classifications such as highways, urban roads, and rural roads; and the road surface point cloud model accurately describes the geometry and spatial structure of the road surface in the form of three-dimensional point cloud data. Simultaneously, the system acquires the vehicle model and load of each vehicle in the current platoon. The vehicle model includes basic parameters such as brand, model, and size specifications; the vehicle load reflects the actual loaded weight of each vehicle, a parameter that directly affects the vehicle's power and braking performance.

[0049] Then, based on the constraints such as road slope, road type, vehicle model in the platoon, and vehicle load, matching vehicle sample data is retrieved from the historical driving database to extract high-frequency speeds and high-frequency distances of the vehicle samples. High-frequency speed refers to the most commonly used speed of historical vehicle samples under similar road and vehicle conditions; high-frequency distance is the most commonly used following distance under the corresponding conditions. Subsequently, based on the retrieved high-frequency speeds and distances of the vehicle samples, and combined with the current number of vehicles in the platoon, the corresponding vehicle point cloud model is retrieved from the vehicle model library. Next, following the path direction from the starting point to the end point, the spatial configuration of the vehicle platoon is deployed on the road point cloud model according to the calculated speed and distance parameters, generating a visual point cloud of the expected platoon posture. This visual point cloud of the expected platoon posture fully describes the ideal spatial relationship and driving posture configuration that each vehicle in the platoon should maintain under specific road conditions.

[0050] By taking into full account the characteristics of the actual road environment and the vehicle's operating characteristics, and by generating a scientific and reasonable formation attitude expectation based on the statistical analysis results of a large amount of historical driving data, the accuracy of formation control and environmental adaptability can be improved, thereby effectively enhancing the safety and stability of vehicle formation driving.

[0051] Further, retrieving high-frequency speeds and high-frequency distances of vehicle samples that satisfy the road gradient, road type, platoon vehicle model, and platoon vehicle load includes:

[0052] S131. Retrieve multiple sets of accident-free driving data of sample vehicles that satisfy the road slope, road type, vehicle model in the platoon, and vehicle load in the platoon. Each set of accident-free driving data of sample vehicles includes a normalized feature value of driving speed and a normalized feature value of driving distance.

[0053] S132. Using the normalized feature value of driving speed and the normalized feature value of driving distance as two-dimensional coordinate axis attributes, the multiple sample vehicle accident-free driving data are distributed to obtain multiple driving data distribution coordinates.

[0054] S133. Extract the concentrated coordinates of the multiple driving data distribution coordinates to generate the high-frequency speed of the vehicle sample and the high-frequency distance of the vehicle sample.

[0055] In a preferred embodiment, firstly, multiple sets of accident-free driving data for sample vehicles are retrieved from a historical driving database, satisfying multiple constraints such as road gradient, road type, vehicle model in the platoon, and vehicle load in the platoon. The accident-free driving data for sample vehicles refers to records of historical vehicles driving safely under the same or similar driving conditions. Each set of accident-free driving data for sample vehicles includes a normalized characteristic value for driving speed and a normalized characteristic value for following distance. The normalized characteristic value for driving speed is a value obtained by normalizing the original driving speed according to a preset standard, and the normalized characteristic value for following distance is a value obtained by normalizing the original following distance according to the same standard.

[0056] Then, a two-dimensional coordinate system is constructed, with normalized driving speed as the horizontal axis attribute and normalized driving distance as the vertical axis attribute. Multiple accident-free driving data sets of sample vehicles are distributed and mapped within this two-dimensional coordinate system, obtaining multiple driving data distribution coordinates. Each driving data distribution coordinate corresponds to the location point of a historical driving record in the two-dimensional parameter space. Subsequently, a data clustering analysis algorithm is used to extract the concentrated coordinates of the multiple driving data distribution coordinates, i.e., the center position of the region with the highest data point density. The horizontal axis values ​​of these concentrated coordinates are inversely normalized to generate the high-frequency speed of the vehicle samples, and the vertical axis values ​​are inversely normalized to generate the high-frequency vehicle distance of the vehicle samples.

[0057] Through the above data mining and statistical analysis, the optimal driving parameters can be accurately extracted from a large amount of historical safe driving data, ensuring the scientific nature and reliability of the formation attitude expectation and improving the safety and practicality of the formation control strategy.

[0058] Furthermore, based on the image information of the first following vehicle and the image information of the first vehicle, formation posture recognition is performed to obtain a visualized point cloud for formation posture monitoring, including:

[0059] S31. Extract the front-facing image library of the first following vehicle model, perform similarity analysis with the image information of the first following vehicle, and obtain a selected front-facing image. The selected front-facing image is stored in the front-facing image library with a predefined first following vehicle posture identifier.

[0060] S32. Extract the rear-direction image library of the first vehicle model, perform similarity analysis with the first vehicle image information, and obtain a selected rear-direction image, wherein the selected rear-direction image is pre-stored in the rear-direction image library with a predefined first vehicle posture identifier.

[0061] S33. Based on the first vehicle attitude identifier and the first following vehicle attitude identifier, locate the formation attitude monitoring visualization point cloud by using the spatial position of the first following vehicle image information and the first vehicle image information.

[0062] In a preferred embodiment, firstly, a front-facing image library corresponding to the model of the first following vehicle is extracted from a pre-built image library. The standard images in this front-facing image library are then compared with the real-time acquired images of the first following vehicle using a similarity analysis. An image matching algorithm is used to obtain the selected front-facing image with the highest similarity. This selected front-facing image contains a predefined first-following vehicle attitude identifier stored in the front-facing image library. This attitude identifier includes key attitude parameters such as the vehicle's spatial orientation angle and body tilt angle. Simultaneously, a rear-facing image library corresponding to the model of the first vehicle is extracted from the pre-built image library. The standard images in this rear-facing image library are then compared with the real-time acquired images of the first vehicle using a similarity analysis. An image matching algorithm is used to obtain the selected rear-facing image with the highest similarity. This selected rear-facing image also contains a predefined first-vehicle attitude identifier stored in the rear-facing image library. This attitude identifier also includes key attitude parameters such as the vehicle's spatial orientation angle and body tilt angle.

[0063] Subsequently, based on the obtained attitude identifiers of the first vehicle and the first following vehicle, and combined with the spatial position information of the first following vehicle image and the first vehicle image, a three-dimensional spatial positioning calculation is performed. Since the acquired images are binocular images, the spatial coordinates of the target vehicle can be calculated using the principle of stereo vision. Combined with the attitude identifier information, the precise spatial attitude of the vehicle is determined, generating a visual point cloud of formation attitude monitoring that includes the actual spatial relationship between the two vehicles.

[0064] Through the aforementioned posture recognition based on image recognition and stereo vision technology, accurate perception and quantitative representation of the real-time posture of platooned vehicles were achieved, improving the accuracy and real-time performance of platoon posture monitoring.

[0065] Furthermore, an image library of the frontal direction of the first following vehicle model is extracted, and a similarity analysis is performed with the image information of the first following vehicle to obtain a selected image of the frontal direction. This process includes:

[0066] S341. Using the model of the first following vehicle, the model of the binocular image acquisition device, and the set acquisition parameters as constraints, record the image acquisition location dataset, the vehicle front direction image dataset, and the vehicle attitude point cloud identifier dataset.

[0067] S342. Using the set of images in the direction of the vehicle's front as the compilation target of the encoder, and using the dataset of image acquisition locations as input variables, train a convolutional neural network to obtain a compiler for the direction of the vehicle's front.

[0068] S343. Configure the head direction acquisition position constraint of the first following vehicle model through the user terminal;

[0069] S344. Based on the vehicle front direction acquisition position constraint, the vehicle front direction compiler is used to encode the image set of the initial vehicle front direction;

[0070] S345. Associate and store the initial vehicle front direction image set with the vehicle attitude point cloud identifier dataset to generate the vehicle front direction image library.

[0071] In a preferred embodiment, before executing step S31, a front-facing image library needs to be pre-built. First, using the model of the first following vehicle, the model of the binocular image acquisition device, and the set acquisition parameters as data acquisition constraints, three types of basic datasets are recorded through actual acquisition methods: an image acquisition location dataset, a front-facing image set, and a vehicle attitude point cloud label dataset. The image acquisition location dataset records the spatial position and angle information of the acquisition device; the front-facing image set contains front-facing images acquired at different positions and angles; and the vehicle attitude point cloud label dataset records the actual attitude parameters of the vehicle in each image.

[0072] Then, a convolutional neural network training framework was constructed, using the set of images in the direction of the vehicle's front as the output of the encoder and the image acquisition location dataset as the input variable of the network. A vehicle front orientation compiler was obtained through a deep learning training process. This compiler can generate vehicle front images from the corresponding viewpoint based on given acquisition location parameters. Simultaneously, through the user interface, the administrator configures the vehicle front orientation acquisition location constraints corresponding to the first following vehicle model. These constraints define key parameters such as the spatial range, angular range, and sampling density of the generated vehicle front images.

[0073] Subsequently, based on the configured front-facing direction acquisition position constraints, a large-scale image encoding and generation process is performed using a trained front-facing direction compiler to obtain an initial front-facing direction image set covering various possible viewpoints and positions, thus achieving intelligent expansion of the limited actual acquisition data. Afterwards, the generated initial front-facing direction image set is associated and stored with the corresponding vehicle attitude point cloud label dataset, establishing a one-to-one correspondence between images and attitude labels, ultimately generating a complete front-facing direction image library.

[0074] The above-mentioned machine learning-based image library construction method overcomes the problem of limited manually collected data, generates an image library of vehicle front direction covering all-round perspectives, and improves the accuracy and robustness of subsequent posture recognition.

[0075] Furthermore, an image library of the frontal direction of the first following vehicle model is extracted, and a similarity analysis is performed with the image information of the first following vehicle to obtain a selected image of the frontal direction, including:

[0076] S311. Extract the first front-facing image from the front-facing image library;

[0077] S312. Extract the first shape feature, first color feature, and first texture feature of the first vehicle head direction image through the first feature extraction channel of the twin neural network;

[0078] S313. Extract the second shape feature, second color feature, and second texture feature of the first following vehicle image information through the second feature extraction channel of the twin neural network;

[0079] S314. Compare the first shape feature and the second shape feature to calculate the shape similarity; compare the first color feature and the second color feature to calculate the color similarity; compare the first texture feature and the second texture feature to calculate the texture similarity.

[0080] S315. Based on the model of the first following vehicle, configure the shape weight distribution, color weight distribution, and texture weight distribution of the vehicle front;

[0081] S316. Based on the shape weight distribution, the color weight distribution, and the texture weight distribution, the shape similarity, the color similarity, and the texture similarity are weighted to obtain a comprehensive similarity.

[0082] S317. Extract the image with the highest overall similarity from the vehicle front direction image library:

[0083] S318. When the maximum value of the comprehensive similarity is greater than or equal to the comprehensive similarity threshold, the image with the maximum comprehensive similarity is set as the selected vehicle head direction image;

[0084] S319. Otherwise, expand the vehicle front direction acquisition position constraint to update the vehicle front direction image library.

[0085] In a preferred embodiment, firstly, front-facing images are extracted sequentially from a pre-built front-facing image library according to a preset traversal order, and denoted as the first front-facing image, which serves as the benchmark comparison image for the current matching loop. The first front-facing image contains complete front-facing visual information and corresponding posture identification data.

[0086] Then, a Siamese neural network is invoked. This Siamese neural network employs a dual-branch symmetrical structure, consisting of a first feature extraction channel and a second feature extraction channel. These two feature extraction channels share the same network parameters and weights, ensuring consistency in feature extraction methods for the input image. Through the first feature extraction channel in the Siamese neural network, deep feature analysis is performed on the first vehicle-facing image to obtain the first shape feature, first color feature, and first texture feature of the first following vehicle image. This first feature extraction channel consists of multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layers use convolutional kernels of different sizes to extract local features and edge information from the image. The pooling layers perform feature dimensionality reduction and spatial invariance processing through max pooling or average pooling operations. The fully connected layers realize the mapping and abstract expression from low-level visual features to high-level semantic features. When extracting the first shape feature, the first feature extraction channel focuses on the vehicle's edge contour, geometric shape, and structural proportion information, quantifying the vehicle's geometric feature vector through edge detection operators and shape descriptors. When extracting the first color feature, the first feature extraction channel analyzes the color distribution histogram, dominant hue, and color saturation information in the HSV and RGB color spaces. When extracting the first texture feature, the first feature extraction channel uses gray-level co-occurrence matrix and local binary mode algorithms to analyze the texture roughness, directionality, and periodicity of the vehicle surface. Simultaneously, the second feature extraction channel in the Siamese neural network performs the same feature extraction process as the first feature extraction channel on the real-time acquired image information of the first following vehicle. Due to the weight-sharing mechanism of the Siamese neural network, the second feature extraction channel uses the exact same network architecture, convolution kernel parameters, activation functions, and inter-layer connections as the first feature extraction channel, ensuring consistency and comparability of feature extraction. The second feature extraction channel processes the image information of the first following vehicle to obtain the second shape feature, the second color feature, and the second texture feature, forming feature vector sets corresponding to the first shape feature, the first color feature, and the first texture feature.

[0087] Subsequently, various similarity calculation methods were employed for feature comparison. Specifically, shape similarity was calculated by comparing the first and second shape features, for example, using Hausdorff distance and shape context matching algorithms to quantify the geometric differences between the two shape feature vectors; color similarity was calculated by comparing the first and second color features, for example, using the Bach coefficient and histogram intersection algorithm to measure the similarity of color distributions; and texture similarity was calculated by comparing the first and second texture features, for example, using mutual information and correlation coefficient methods to evaluate the matching degree of texture patterns. Each similarity calculation result was normalized to the [0,1] interval, with values ​​closer to 1 indicating higher similarity.

[0088] Subsequently, based on the statistical analysis of the visual feature importance of the first following vehicle model, and through historical recognition accuracy data and feature saliency analysis, the weight distribution of the three types of features was dynamically configured. For models with obvious outlines, the shape weight distribution was set to a higher value; for models with prominent color features, the color weight distribution received a larger proportion; and for models with rich surface textures, the texture weight distribution was increased accordingly. The weight distribution values ​​met the normalization constraint, ensuring that the sum of the three weights equaled 1. Then, a weighted average fusion algorithm was used to linearly combine shape similarity, color similarity, and texture similarity according to their corresponding weight distributions to obtain the comprehensive similarity. The calculation formula is: Comprehensive Similarity = Shape Weight Distribution × Shape Similarity + Color Weight Distribution × Color Similarity + Texture Weight Distribution × Texture Similarity. This comprehensive similarity reflects the overall matching degree of the two images across multiple feature dimensions.

[0089] Then, a traversal mechanism is established to repeatedly execute steps S311 to S316 of the similarity calculation process for each image in the vehicle-facing image library, creating a comprehensive similarity value list. A numerical comparison algorithm is used to extract the maximum value in the list and its corresponding image, yielding the maximum comprehensive similarity value and the image with the highest comprehensive similarity. Next, the calculated maximum comprehensive similarity value is compared with a preset comprehensive similarity threshold. This threshold is set based on vehicle recognition accuracy requirements and system performance balance, typically between 0.8 and 0.95. When the maximum comprehensive similarity value is greater than the threshold, a successful match is confirmed, and the corresponding image is designated as the selected vehicle-facing image. When the maximum comprehensive similarity value is less than the threshold, it indicates that the current coverage of the vehicle-facing image library is insufficient to support accurate matching. In this case, an image library expansion mechanism is automatically triggered, expanding the angle range of the vehicle-facing acquisition position constraint by 10-20 degrees and the distance range by 0.5-1 meter. The vehicle-facing compiler is then invoked to generate new image data, updating the vehicle-facing image library before the matching process is re-executed.

[0090] Through the above steps, high-precision vehicle image recognition and matching were achieved, improving the accuracy and adaptability of formation posture recognition.

[0091] Furthermore, based on the first following vehicle model, shape weight distribution, color weight distribution, and texture weight distribution are configured, including:

[0092] S3151. Cluster the shape features of the N directions of the front images of the first following vehicle model to obtain the shape feature clustering results. Calculate the ratio of the number of clusters in the shape feature clustering results to N, and set it as the shape weight reference factor. The shape weight distribution is positively correlated with the shape weight reference factor. N is an integer, and N≥50.

[0093] S3152. Cluster the color features of the N directions of the front images of the first following vehicle model to obtain the color feature clustering results. Calculate the ratio of the number of clusters in the color feature clustering results to N, and set it as the color weight reference factor. The color weight distribution is positively correlated with the color weight reference factor.

[0094] S3153. Cluster the texture features of the N-direction front images of the first following vehicle model to obtain the texture feature clustering results. Calculate the ratio of the number of clusters in the texture feature clustering results to N, and set it as the texture weight reference factor. The texture weight distribution is positively correlated with the texture weight reference factor.

[0095] S3154. Configure the shape weight distribution, the color weight distribution, and the texture weight distribution based on the shape weight reference factor, the color weight reference factor, and the texture weight reference factor.

[0096] In a preferred embodiment, firstly, N front-end images corresponding to the first following vehicle model are extracted from a pre-built vehicle image database, where N is a positive integer not less than 50 to ensure the statistical validity of the sample size. Shape feature vectors are extracted from each front-end image, including feature parameters such as edge contours, geometric shapes, and structural proportions, forming a shape feature dataset. Then, the K-means clustering algorithm is used to perform cluster analysis on the shape feature dataset. By iteratively optimizing cluster centers and minimizing intra-cluster distances, the shape feature clustering results are obtained. During the clustering process, the elbow rule and contour coefficient are used to evaluate the optimal number of clusters to ensure the rationality and stability of the clustering results. The final number of clusters in the shape feature clustering results is counted, and the ratio of this number of clusters to the total number of samples N is calculated. This ratio is defined as the shape weight reference factor. The shape weight reference factor reflects the diversity of the vehicle model's shape features; a larger ratio indicates richer shape variations and higher importance of shape features in recognition. Therefore, the shape weight distribution is positively correlated with the shape weight reference factor.

[0097] Simultaneously, color features were extracted from the front images of the first following vehicle model from N directions, obtaining a set of color feature vectors containing HSV and RGB color space information. A Gaussian mixture model-based clustering algorithm was used to perform cluster analysis on the color feature data, processing the probability distribution characteristics and multimodal features of the color features. The clustering parameters were iteratively optimized using the expectation-maximization algorithm to obtain the color feature clustering results. The optimal number of clusters was evaluated using the Bayesian information criterion and the Akaike information criterion to ensure that the clustering results accurately reflect the color feature distribution pattern of the vehicle model. The number of clusters in the color feature clustering results was counted, and the ratio of the number of clusters to N was calculated and set as the color weight reference factor. The color weight reference factor quantifies the discriminative power and variation range of the color features of the vehicle model; a larger ratio indicates more discriminative color features, and correspondingly, the color weight distribution is positively correlated with the color weight reference factor.

[0098] Furthermore, texture features were extracted from the front images of the first following vehicle model from N directions. A set of texture feature descriptors was obtained using the gray-level co-occurrence matrix and local binary mode algorithms. Spectral clustering was used to perform cluster analysis on the texture feature data. Constructing a similarity graph and eigenvalue decomposition effectively handled the nonlinear distribution characteristics of the texture features. The optimal number of clusters was determined using the eigenvalue gap criterion and modularity optimization method, obtaining the texture feature clustering results. The final number of clusters for texture feature clustering was counted, and the ratio of the number of clusters to the total number of samples N was calculated. This ratio was defined as the texture weight reference factor. The texture weight reference factor characterizes the complexity and diversity of the surface texture of the vehicle model. A larger ratio indicates richer texture features and stronger recognition ability; therefore, the texture weight distribution is positively correlated with the texture weight reference factor.

[0099] Subsequently, based on the calculated shape weight reference factor, color weight reference factor, and texture weight reference factor, a normalized weight allocation algorithm is used to configure the final shape weight distribution, color weight distribution, and texture weight distribution. The specific calculation process is as follows: first, the sum of the three weight reference factors is calculated; then, each weight reference factor is divided by the sum to obtain the initial weight ratio. To avoid extreme weight distributions, a weight smoothing mechanism is introduced to adjust the initial weight ratios, ensuring that any weight value is neither lower than 0.1 nor higher than 0.7. Finally, a softmax normalization function is used to ensure that the sum of the shape weight distribution, color weight distribution, and texture weight distribution is strictly equal to 1, satisfying the probability distribution constraints.

[0100] By using the weight configuration based on feature clustering analysis and statistical learning, the importance weights of each feature dimension can be adaptively adjusted according to the visual feature distribution characteristics of different vehicle models, thereby improving the accuracy and relevance of vehicle recognition.

[0101] Example 2, please refer toFigure 2 , Figure 2 A schematic diagram illustrating an embodiment of the electronic device provided in this application. For example... Figure 2 As shown, an electronic device 100 provided in this application embodiment includes a memory 110, a processor 120, and a computer software program 111 stored in the memory 110 and executable on the processor 120. When the processor 120 executes the computer software program 111, it implements the vehicle platooning visualization and interactive adjustment method provided in Embodiment 1.

[0102] Example 3, please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided in this application. For example... Figure 3 As shown, this embodiment provides a computer-readable storage medium 200, on which a computer software program 111 is stored. When the computer software program 111 is executed by a processor, it implements the vehicle platooning visualization and interactive adjustment method provided in Embodiment 1.

[0103] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0104] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Although preferred embodiments of this application have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0109] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for visually and interactively adjusting vehicle platooning, characterized in that, Applied to the first edge node embedded in the first following vehicle, including: The navigator receives the expected visualization point cloud of the formation attitude of the first vehicle and the first following vehicle, wherein the first following vehicle is the vehicle following the first vehicle, and the first vehicle is the navigator or the following vehicle. Image information of the first following vehicle is acquired from the binocular image acquisition device behind the first vehicle, and image information of the first vehicle is acquired from the binocular image acquisition device in front of the first following vehicle. Based on the image information of the first following vehicle and the image information of the first vehicle, formation posture recognition is performed to obtain a visual point cloud for formation posture monitoring. Based on the difference point cloud information between the desired formation attitude visualization point cloud and the formation attitude monitoring visualization point cloud, the formation attitude monitoring visualization point cloud is rendered and marked and sent to the lead vehicle through the V2V communication network to perform vehicle formation optimization. Specifically, based on the image information of the first following vehicle and the image information of the first vehicle, formation posture recognition is performed to obtain a visualized point cloud for formation posture monitoring, including: Extract the front-facing image library of the first following vehicle model, perform similarity analysis with the image information of the first following vehicle, and obtain a selected front-facing image. The selected front-facing image is stored in the front-facing image library with a predefined first following vehicle posture identifier. Extract the rear-direction image library of the first vehicle model, perform similarity analysis with the image information of the first vehicle, and obtain the selected rear-direction image. The selected rear-direction image is stored in the rear-direction image library with a predefined first vehicle posture identifier. Based on the first vehicle attitude identifier and the first following vehicle attitude identifier, the spatial location of the first following vehicle image information and the first vehicle image information is used to obtain the visual point cloud of the formation attitude monitoring. The process includes extracting an image library of the frontal direction of the first following vehicle model, performing similarity analysis with the image information of the first following vehicle, and obtaining a selected image of the frontal direction. This process previously includes: With the model of the first following vehicle, the model of the binocular image acquisition device, and the set acquisition parameters as constraints, record the image acquisition location dataset, the vehicle front direction image dataset, and the vehicle attitude point cloud label dataset. Using the set of images in the direction of the vehicle's front as the compilation target of the encoder, and using the dataset of image acquisition locations as input variables, a convolutional neural network is trained to obtain the compiler for the direction of the vehicle's front. Configure the head direction acquisition position constraint of the first following vehicle model through the user terminal; Based on the vehicle front direction acquisition position constraint, the vehicle front direction compiler is used to encode the image to obtain an initial vehicle front direction image set; The initial front-facing image set is associated and stored with the vehicle attitude point cloud identifier dataset to generate the front-facing image library.

2. The method as described in claim 1, characterized in that, The desired visualization of the formation attitude of the lead vehicle and the first following vehicle is received from the lead vehicle, including: The target road features are downloaded from the cloud map using a navigation vehicle. These features include road gradient, road type, and road surface point cloud model. Obtain the vehicle types and loads in the formation; Retrieve high-frequency speeds and high-frequency distances of vehicle samples that satisfy the road gradient, road type, platoon vehicle model, and platoon vehicle load; Based on the high-frequency speed and high-frequency distance of the vehicle samples, combined with the number of vehicles in the formation, the vehicle point cloud model is retrieved, and the vehicle formation is deployed on the road point cloud model from the starting point to the ending point to obtain the visual point cloud of the formation attitude expectation.

3. The method as described in claim 2, characterized in that, Retrieving high-frequency speeds and high-frequency distances of vehicle samples that satisfy the road gradient, road type, platoon vehicle model, and platoon vehicle load includes: Retrieve multiple sets of accident-free driving data for sample vehicles that satisfy the road gradient, road type, vehicle model in the platoon, and vehicle load in the platoon. Each set of accident-free driving data for sample vehicles includes a normalized feature value of driving speed and a normalized feature value of driving distance. Using the normalized feature value of driving speed and the normalized feature value of driving distance as two-dimensional coordinate axis attributes, the multiple sample vehicle accident-free driving data are distributed to obtain multiple driving data distribution coordinates; Extract the concentrated coordinates of the multiple driving data distribution coordinates to generate the high-frequency speed of the vehicle sample and the high-frequency distance between the vehicle samples.

4. The method as described in claim 1, characterized in that, Extract the frontal view image library of the first following vehicle model, perform similarity analysis with the image information of the first following vehicle, and obtain the selected frontal view image, including: Extract the first frontal image from the frontal image library; The first shape feature, first color feature, and first texture feature of the first vehicle front direction image are extracted through the first feature extraction channel of the twin neural network; The second shape feature, second color feature, and second texture feature of the first following vehicle image information are extracted through the second feature extraction channel of the twin neural network; Compare the first shape feature and the second shape feature to calculate the shape similarity; compare the first color feature and the second color feature to calculate the color similarity; compare the first texture feature and the second texture feature to calculate the texture similarity. Based on the model of the first following vehicle, configure the shape weight distribution, color weight distribution, and texture weight distribution of the vehicle's front end; Based on the shape weight distribution, the color weight distribution, and the texture weight distribution, the shape similarity, the color similarity, and the texture similarity are weighted to obtain a comprehensive similarity. Extract the image with the highest overall similarity from the image library of vehicle front view: When the maximum value of the comprehensive similarity is greater than or equal to the comprehensive similarity threshold, the image with the maximum comprehensive similarity is set as the selected vehicle head direction image; Otherwise, expand the constraints on the acquisition position in the vehicle's front direction to update the image library in the vehicle's front direction.

5. The method as described in claim 4, characterized in that, Based on the first following vehicle model, configure shape weight distribution, color weight distribution, and texture weight distribution, including: Cluster the shape features of the N directions of the front images of the first following vehicle model to obtain the shape feature clustering results. Calculate the ratio of the number of clusters in the shape feature clustering results to N, and set it as the shape weight reference factor. The shape weight distribution is positively correlated with the shape weight reference factor. N is an integer, and N≥50. Cluster the color features of the N directions of the front images of the first following vehicle model to obtain color feature clustering results. Calculate the ratio of the number of clusters in the color feature clustering results to N, and set it as the color weight reference factor. The color weight distribution is positively correlated with the color weight reference factor. Cluster the texture features of the N-direction front images of the first following vehicle model to obtain the texture feature clustering results. Calculate the ratio of the number of clusters in the texture feature clustering results to N, and set it as the texture weight reference factor. The texture weight distribution is positively correlated with the texture weight reference factor. The shape weight distribution, color weight distribution, and texture weight distribution are configured based on the shape weight reference factor, the color weight reference factor, and the texture weight reference factor.

6. An electronic device, characterized in that, include: Memory, used to store computer software programs; A processor for reading and executing the computer software program to implement the method as described in any one of claims 1 to 5.

7. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images