Semantic detection-oriented resource allocation method in vehicle queue mode

Abstract

The invention discloses a semantic detection-oriented resource allocation method in a vehicle queue mode, which comprises the following steps of: constructing a mathematical relationship model between video semantic detection accuracy and available wireless resource quantity under the conditions that a PM and a PL carry out V2V communication and the PM and a roadside base station carry out V2I communication, and constructing a resource allocation optimization problem on the basis of the mathematical relationship model; converting a wireless link selection problem into an equivalence problem about PM communication mode selection, converting the equivalence problem into a maximum weight matching problem of a bipartite graph, and solving by using a Hungary algorithm to obtain a value of a binary variable of PM communication mode selection; a resource allocation optimization problem is converted into a convex optimization problem about the bandwidth and the transmitting power, an estimated value of a PM communication mode selection variable is combined, a convex optimization algorithm is used for solving, and the bandwidth allocated to the PM and the transmitting power of the PM are obtained; the method has the advantages that the V2X vehicle-road cooperation performance in the vehicle queue is improved, video semantic detection is carried out, and the utilization efficiency of wireless resources is improved.

Description

Technical Field

[0001] The present invention relates to the intersection of intelligent transportation systems and wireless communication technologies, and in particular to a resource allocation method for semantic detection in a vehicle platoon mode, which aims to improve the V2X (Vehicle-to-Everything) vehicle-road collaboration performance in a vehicle platoon. Background Art

[0002] In recent years, the research enthusiasm of academia and industry for vehicle-to-everything (V2X) communication technology has continued to rise. V2X covers a variety of interaction forms such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I), and is the core force driving the evolution of intelligent transportation systems (ITS). The core of realizing intelligent transportation lies in whether vehicles can quickly perceive and instantly exchange surrounding environment information with other vehicles. This requirement is particularly urgent in autonomous driving scenarios, because autonomous driving vehicles must carry out large-scale coordinated operations with vehicles equipped with artificial intelligence (AI) capabilities and the Internet of Vehicles (IoV) built on V2X communication. In addition, communication devices are gradually evolving from simple data transceivers to intelligent entities with computing capabilities.

[0003] As an emerging autonomous driving mode, vehicle platooning aims to address urban and highway traffic congestion by optimizing traffic organization and resource utilization efficiency. Vehicles in a vehicle platoon use V2X communication to achieve driving information interaction and collaborative driving to enhance the safety of the traffic system. In a vehicle platoon, the platoon leader (PL) is responsible for unified command of the platoon's driving speed and driving path, while the platoon member (PM) strictly follows the prescribed distance. Given that cameras can collect more than 90% of road environment information, each PM is equipped with multiple high-resolution cameras to achieve all-round visual coverage. In order to achieve environmental perception beyond the line of sight of a single vehicle, PM needs to transmit the captured video data information to PL in real time. This video data information is crucial for PL to accurately judge and execute driving actions such as turning, deceleration, and lane changing.

[0004] The process of PM transmitting video data information to PL in real time will bring great pressure to the limited wireless resources, so resource allocation is needed to make more reasonable use of limited wireless resources. V2X communication in the context of autonomous driving not only focuses on the transmission efficiency of information, but also emphasizes the accurate grasp of information content, which is consistent with the concept of semantic communication. However, the existing resource allocation mechanism generally ignores the semantic level of information, resulting in low efficiency of wireless resource utilization. For the existing resource allocation mechanism, most of the current work only focuses on V2I, that is, vehicles tend to use vehicle networking technology to further offload computing tasks to edge computing (EC) servers with rich computing resources. Such servers are generally deployed at base stations or roadside units and are responsible for performing collaborative computing tasks. However, in the vehicle platoon mode, focusing only on V2I will lead to a decrease in the utilization of wireless resources. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a resource allocation method for semantic detection in a vehicle platoon mode, which can use the Internet of Vehicles technology to offload computing tasks to an edge computing server with rich computing resources for collaborative computing. On this basis, a wireless communication link is established between PM and PL, thereby improving the V2X vehicle-road collaborative performance in the vehicle platoon, performing video semantic detection, and improving the utilization efficiency of wireless resources.

[0006] The technical solution adopted by the present invention to solve the above technical problem is: a resource allocation method for semantic detection in a vehicle queue mode, characterized by comprising the following steps:

[0007] Step 1: Build a video semantic communication vehicle networking system model covering PL, PM, and roadside base stations equipped with EC servers:

[0008] 1) Set up roadside base stations equipped with EC servers: Divide the highway into multiple consecutive sections, and set up a roadside base station equipped with an EC server in each section. The network signal of each roadside base station completely covers the section where it is located, and can communicate with any vehicle traveling in the section where it is located;

[0009] 2) Set up a vehicle queue: The vehicle queue consists of 1 PL and M PMs, a total of M+1 vehicles, each PM is equipped with a high-definition camera; the time domain of the vehicle queue when it is running continuously is divided into multiple time slots, and all vehicles in the vehicle queue in a time slot are set to travel in the same interval; where M≥1;

[0010] 3) Clarify the two tasks of the video semantic communication vehicle network system model, namely PM communication mode selection and wireless resource allocation scheme determination;

[0011] In the video semantic communication vehicle network system model, each PM uses a high-definition camera to capture road information video in real time, and the PL and the roadside base station obtain their respective channel status data in real time. Then, the roadside base station selects the communication link with the relatively best channel status as the optimal communication link for each PM from the V2V communication link between the PM and the PL and the V2I communication link between the PM and the roadside base station based on its own channel status data and the channel status data of the PL, and considering the processing capacity limitation of the PL. Then, a node in the PL or roadside base station corresponding to the optimal communication link is determined as the receiving node. The PM encodes and compresses the road information video according to the wireless resource allocation scheme, and transmits the encoded road information video to the receiving node through the optimal communication link. The receiving node decodes and restores the received road information video, and performs semantic analysis on the decoded and restored road information video. The semantic analysis results are summarized to the PL to guide the overall driving behavior of the vehicle queue.

[0012] Step 2: Construct a mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the mth PM and PL perform V2V communication in the tth time slot, which is described as: The mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the mth PM performs V2I communication with the roadside base station in the tth time slot is constructed, which is described as: Where, m = 1, 2, ..., M, f() is the function representation, It represents the accuracy of video semantic detection when the m-th PM and PL perform V2V communication in the t-th time slot. The function of It represents the accuracy of video semantic detection when the mth PM communicates with the roadside base station in the tth time slot. The function of represents the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot, represents the transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot, and They are calculated based on the bandwidth allocated to the mth PM and the transmission power of the mth PM in the tth time slot. The bandwidth and transmission power are the available wireless resources. α1, α2, α3, η1, η2 are all function parameters, and e is a natural constant.

[0013] Step 3: Construct a resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system, which is described as:

[0014]

[0015] Among them, l m represents the binary variable for the communication mode selection of the mth PM in the tth time slot, l m =1 means that the V2V communication link between the mth PM and the PL is selected for the mth PM in the tth time slot, l m = 0 means that the V2I communication link between the mth PM and the roadside base station is selected for the mth PM in the tth time slot. m =1 when f(R m )for When l m = 0 when f(R m )for R m =B m log2(1+γ m ), when l m =1 when R m for When l m = 0 when R m for B m represents the bandwidth allocated to the mth PM in the tth time slot. m =1 when γ m for represents the received signal-to-noise ratio of PL when the mth PM and PL perform V2V communication in the tth time slot. m = 0 when γ m for represents the received signal-to-noise ratio of the roadside base station when the mth PM communicates with the roadside base station in the tth time slot, P m represents the transmission power of the mth PM in the tth time slot, represents the maximum transmission power of the mth PM, represents the maximum bandwidth that can be used by all PMs in V2V communication with PL, It represents the maximum bandwidth that can be used by all PMs for V2I communication with roadside base stations. represents the maximum time for the mth PM to transmit the compressed and coded road information video, γ th represents the preset signal-to-noise ratio threshold, N represents the number of nodes in PL, w1, w2, w3, and w4 are all weights, and their values ​​are all negative;

[0016] Step 4: Based on the resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system, PM communication mode selection and wireless resource allocation scheme design are performed as follows:

[0017] Step 4.1: Describe the sub-problem of PM communication mode selection in the resource allocation optimization problem between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system in the tth time slot as follows:

[0018]

[0019] Among them, h m,B represents the channel gain of the V2I communication link between the mth PM and the roadside base station in the tth time slot, h m,PL represents the channel gain of the V2V communication link between the mth PM and PL in the tth time slot;

[0020] Step 4.2: Convert the subproblem about PM communication mode selection into the maximum weight matching problem of bipartite graph, and then use the Hungarian algorithm to solve the maximum weight matching problem of bipartite graph to obtain an initial result set; then count the number of binary values ​​1 in the initial result set. If the counted number is less than or equal to N, directly use the initial result set as the final result set; if the counted number is greater than N, sort the weights corresponding to all binary values ​​1 in the initial result set from large to small, and modify the binary values ​​1 corresponding to the N+1th to the last weight in all the sorted weights to binary value 0 to obtain the final result set; then use each binary value in the final result set as the value of the binary variable for communication mode selection of each PM in the tth time slot, and obtain l m The value of Complete the selection of PM communication mode;

[0021] Step 4.3: The resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system is transformed into a convex optimization problem, which is described as:

[0022]

[0023] Step 4.4: Use the convex optimization algorithm to solve the convex optimization problem, obtain the bandwidth allocated to each PM in the tth time slot and the transmission power of each PM in the tth time slot, complete the design of the wireless resource allocation plan for each PM in the tth time slot, and output the wireless resource allocation plan for each PM in the tth time slot, as well as the objective function value of the wireless resource allocation plan for all PMs in the tth time slot.

[0024] In step 1, the duration of each time slot is 10 milliseconds.

[0025] The specific process of step 2 is as follows:

[0026] Step 2.1: Calculate the channel gain h of the V2V communication link between the mth PM and PL in the tth time slot m,PL , And calculate the channel gain h of the V2I communication link between the mth PM and the roadside base station in the tth time slot m,B , Among them, g m,PL represents the small-scale Rician fading with unit mean corresponding to the V2V communication link between the mth PM and PL in the tth time slot, g m,B represents the small-scale Rician fading with unit mean corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, β m,PL represents the shadow fading of the log-normal distribution corresponding to the V2V communication link between the mth PM and PL in the tth time slot, β m,B represents the shadow fading of the log-normal distribution corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, A represents the path loss constant, γ represents the path loss exponent, and L m,PL represents the distance between the mth PM and PL in the tth time slot, L m,B represents the distance between the mth PM and the roadside base station in the tth time slot;

[0027] Step 2.2: Assume that LTE-V2X technology is used to implement V2V communication and V2I communication. The received signal-to-noise ratio of PL when the m-th PM and PL perform V2V communication in the t-th time slot is expressed as The received signal-to-noise ratio of the roadside base station when the mth PM performs V2I communication with the roadside base station in the tth time slot is expressed as Among them, P m represents the transmission power of the mth PM in the tth time slot, σ 2 represents the noise power density of additive white Gaussian noise at the receiver, where the receiver is a PL or a roadside base station, B m represents the bandwidth allocated to the mth PM in the tth time slot;

[0028] Step 2.3: Express the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot as The transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot is expressed as

[0029] Step 2.4: The QP value when the mth PM and PL perform V2V communication in the tth time slot The relationship model is expressed as: Will The QP value when the mth PM communicates with the roadside base station in the tth time slot The relationship model is expressed as:

[0030] Step 2.5: The video semantic detection accuracy of the case where the m-th PM and PL perform V2V communication in the t-th time slot is about The function is expressed as Combined with get The accuracy of video semantic detection in the case of V2I communication between the mth PM and the roadside base station in the tth time slot is about The function is expressed as Combined with get

[0031] In step 4.4, the convex optimization algorithm selects the convex optimization toolkit included in Matlab R2016b.

[0032] Compared with the prior art, the advantages of the present invention are:

[0033] 1) The method proposed in the present invention can use the Internet of Vehicles technology to offload computing tasks to edge computing servers with rich computing resources for collaborative computing. On this basis, a wireless communication link is established between PM and PL. At the same time, the wireless link of PM is matched based on the channel state as the selection basis of the wireless communication link, which improves the V2X vehicle-road collaborative performance in the vehicle queue, and performs video semantic detection to improve the utilization efficiency of wireless resources.

[0034] 2) The method of the present invention not only constructs a mathematical relationship model between the accuracy of video semantic detection and the amount of available wireless resources in the wireless communication system, but also further designs a feasible solution framework specifically for the problem model based on the resource allocation optimization problem between the accuracy of video semantic detection and the amount of available wireless resources in the wireless communication system.

[0035] 3) Since the resource allocation optimization problem proposed in the method of the present invention has NP difficulty, the method of the present invention proposes a suboptimal solution, which divides the resource allocation optimization problem into two stages. First, the wireless communication link allocation of each PM is obtained by a bipartite graph-based method based on the channel state. Then, the communication resources allocated to the PM are adjusted based on the convex optimization algorithm to maximize the semantic detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of the overall implementation process of the method of the present invention;

[0037] Figure 2 A schematic diagram of a video semantic communication vehicle networking system model (when the number of roadside base stations is 1) constructed by the method of the present invention;

[0038] Figure 3 is a schematic diagram of a weighted bipartite graph in the method of the present invention;

[0039] Figure 4 A comparison chart of the video semantic detection accuracy of each PM when the vehicle queue travels 500 meters from the starting point and the method of the present invention is used and the existing resource allocation mechanism that only focuses on V2I is used;

[0040] Figure 5 In order to consider the vehicle queue as a whole, when the last PM, PM9, sets the starting distance to 0 meters, 100 meters, 200 meters, 300 meters, 400 meters and 500 meters, a comparison chart of the average video semantic detection accuracy of all PMs in the vehicle queue when the method of the present invention and the existing resource allocation mechanism that only focuses on V2I are respectively used. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below with reference to the accompanying drawings.

[0042] The present invention proposes a resource allocation method for semantic detection in a vehicle platoon mode, and its overall implementation flow chart is as follows: Figure 1 As shown, it includes the following steps:

[0043] Step 1: Build a video semantic communication vehicle network system model covering PL (Platoon Leader), PM (Platoon Member) and roadside base stations equipped with EC (Edge Computing) servers, such as Figure 2 As shown:

[0044] 1) Set up roadside base stations equipped with EC servers: Consider a straight highway, which is divided into multiple continuous sections. A roadside base station equipped with an EC server is set up in each section. The network signal of each roadside base station completely covers the section where it is located and can communicate with any vehicle traveling in the section where it is located.

[0045] Here, there is no strict requirement for the specific location of the roadside base station in the section, but the network signal coverage range of the roadside base station is required to be able to fully include the entire section where it is located, that is, it can communicate with any vehicle traveling in the section where it is located, thereby achieving comprehensive network coverage of each section and seamless connection between sections.

[0046] 2) Set up a vehicle queue: The vehicle queue consists of 1 PL and M PMs, a total of M+1 vehicles. Each PM is equipped with a high-definition camera. In the vehicle queue, each PM receives collaborative perception information from the PL and adjusts its motion state accordingly. Therefore, it is necessary to regularly update the collaborative perception information from the PL. Although the vehicle queue runs continuously, it can be regarded as a discrete time system because of the need for regular updates and motion state control. The time domain of the vehicle queue when it runs continuously is divided into multiple time slots. All vehicles in the vehicle queue in a time slot are set to travel in the same interval. Among them, M≥1, and the duration of each time slot is 10 milliseconds.

[0047] 3) Clarify the two tasks of the video semantic communication vehicle network system model, namely PM communication mode selection and wireless resource allocation scheme determination.

[0048] In the video semantic communication vehicle network system model, each PM uses a high-definition camera to capture road information video in real time, and the PL and the roadside base station obtain their respective channel status data in real time; then the roadside base station selects the communication link with the relatively optimal channel status as the optimal communication link for each PM from the V2V communication link between the PM and the PL and the V2I communication link between the PM and the roadside base station based on its own channel status data and the channel status data of the PL, and considering the processing capacity limitation of the PL, and then determines the PL or a node in the roadside base station corresponding to the optimal communication link as the receiving node. The PM encodes and compresses the road information video according to the wireless resource allocation scheme, and transmits the encoded road information video to the receiving node through the optimal communication link. The receiving node decodes and restores the received road information video, and performs semantic analysis on the decoded and restored road information video, and summarizes the semantic analysis results to the PL to guide the overall driving behavior of the vehicle queue.

[0049] Here, the content of the road information video mainly includes the road conditions and the remaining vehicles outside the vehicle queue; the processing capacity limitation of PL refers to the node number limitation of PL; each PM has a V2V communication link with PL and a V2I communication link with the roadside base station, so each PM has two communication links, and a communication link with a relatively optimal channel state is selected from these two communication links; the compression degree of the encoding compression of the road information video is determined by the wireless resource allocation scheme; semantic analysis is a prior art.

[0050] Step 2: Construct a mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the mth PM and PL perform V2V communication in the tth time slot, which is described as: The mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the mth PM performs V2I communication with the roadside base station in the tth time slot is constructed, which is described as: Where, m = 1, 2, ..., M, f() is the function representation, It represents the accuracy of video semantic detection when the m-th PM and PL perform V2V communication in the t-th time slot. The function of It represents the accuracy of video semantic detection when the mth PM communicates with the roadside base station in the tth time slot. The function of represents the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot, represents the transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot, and They are all calculated based on the bandwidth allocated to the mth PM and the transmission power of the mth PM in the tth time slot. The bandwidth and transmission power are the available wireless resources. α1, α2, α3, η1, η2 are all function parameters. e is a natural constant, e=2.71…

[0051] In this specific embodiment, the specific process of step 2 is:

[0052] Step 2.1: Calculate the channel gain h of the V2V communication link between the mth PM and PL in the tth time slot m,PL , And calculate the channel gain h of the V2I communication link between the mth PM and the roadside base station in the tth time slot m,B , Among them, g m,PL represents the small-scale Rician fading with unit mean corresponding to the V2V communication link between the mth PM and PL in the tth time slot, g m,B represents the small-scale Rician fading with unit mean corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, β m,PL represents the shadow fading of the log-normal distribution corresponding to the V2V communication link between the mth PM and PL in the tth time slot, β m,B represents the shadow fading of the log-normal distribution corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, A represents the path loss constant, γ represents the path loss exponent, and L m,PL represents the distance between the mth PM and PL in the tth time slot, L m,B It represents the distance between the mth PM and the roadside base station in the tth time slot.

[0053] Step 2.2: Assume that LTE-V2X technology is used to implement V2V communication and V2I communication. For this purpose, orthogonal sub-band channels are allocated to the vehicles in the vehicle queue, thereby effectively avoiding co-channel interference. Under this premise, the received signal-to-noise ratio of PL when the m-th PM and PL perform V2V communication in the t-th time slot is expressed as The received signal-to-noise ratio of the roadside base station when the mth PM performs V2I communication with the roadside base station in the tth time slot is expressed as Among them, P m represents the transmission power of the mth PM in the tth time slot, σ 2 represents the noise power density of the additive white Gaussian noise (AWGN) at the receiver, where the receiver is a PL or a roadside base station, B m Represents the bandwidth allocated to the mth PM in the tth time slot.

[0054] Step 2.3: Express the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot as The transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot is expressed as

[0055] Step 2.4: In order to quantify the relationship between the accuracy of video semantic detection and influencing factors such as bandwidth and transmission power, the present invention selects the quantization parameter (Quantization Parameter, QP) in the High Efficiency Video Coding (HEVC) scheme as an intermediary variable to explore its correlation with the transmission rate and the accuracy of video semantic detection. HEVC is an advanced video compression standard and the successor of H.264 / AVC (Advanced Video Coding). Compared with the existing standards, the formulation of the HEVC standard aims to significantly improve the compression performance, and strive to maintain or improve the video quality, so that the bit rate can be reduced by 50% under the same subjective video quality perception. In the video compression coding process, QP, as a key parameter for regulating the balance between coding efficiency and video quality, determines the strength of the quantization process. Quantization is a key step in video compression, which reduces the number of bits required by reducing the precision of image data. Different QP values ​​can intuitively reflect the quality of the encoded video, especially when performing object recognition. Therefore, choosing an appropriate QP to balance video compression and quality is crucial to ensure detection accuracy, which is consistent with the goal of maximizing the accuracy of video semantic detection in the object detection task while ensuring that all vehicles and roadside base stations use radio resources effectively.

[0056] Will The QP value when the mth PM and PL perform V2V communication in the tth time slot The relationship model is expressed as: Will The QP value when the mth PM communicates with the roadside base station in the tth time slot The relationship model is expressed as:

[0057] Step 2.5: The video semantic detection accuracy of the case where the m-th PM and PL perform V2V communication in the t-th time slot is about The function is expressed as Combined with get The accuracy of video semantic detection in the case of V2I communication between the mth PM and the roadside base station in the tth time slot is about The function is expressed as Combined with get

[0058] Step 3: Construct a resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available radio resources in the wireless communication system. The purpose of the present invention is to maximize the accuracy of video semantic detection to optimize the understanding of video semantics while minimizing the cost of radio resources. This involves a trade-off between the accuracy of video semantic detection and the use of radio resources. Therefore, the resource allocation optimization problem is described as:

[0059]

[0060] Among them, max is the maximum value function, st means "subject to...", l m represents the binary variable for the communication mode selection of the mth PM in the tth time slot, l m =1 means that the V2V communication link between the mth PM and the PL is selected for the mth PM in the tth time slot, l m = 0 means that the V2I communication link between the mth PM and the roadside base station is selected for the mth PM in the tth time slot. m =1 when f(R m )for When l m = 0 when f(R m )for R m =B m log2(1+γ m ), when l m =1 when Rm for When l m = 0 when R m for B m represents the bandwidth allocated to the mth PM in the tth time slot. m =1 when γ m for represents the received signal-to-noise ratio of PL when the mth PM and PL perform V2V communication in the tth time slot. m = 0 when γ m for represents the received signal-to-noise ratio of the roadside base station when the mth PM communicates with the roadside base station in the tth time slot, P m represents the transmission power of the mth PM in the tth time slot, represents the maximum transmission power of the mth PM, represents the maximum bandwidth that can be used by all PMs in V2V communication with PL, It represents the maximum bandwidth that can be used by all PMs for V2I communication with roadside base stations. represents the maximum time for the mth PM to transmit the compressed and coded road information video, γ th represents the preset signal-to-noise ratio threshold, N represents the number of nodes in PL, w1, w2, w3, and w4 are all weights, and their values ​​are all negative. After normalization, the weight of each communication cost in the entire communication process is expressed as w i In general, w i A negative value indicates that the higher the accuracy of video semantic detection, the higher the cost. In the resource allocation optimization problem, constraint C1 is used to ensure that the allocated bandwidth and transmit power are not less than the minimum value. Constraint C2 is used to ensure that the total bandwidth of all PMs performing V2I communication does not exceed the maximum bandwidth that all PMs performing V2I communication can use. Constraint C3 is used to ensure that the total bandwidth of all PMs performing V2V communication does not exceed the maximum bandwidth that all PMs performing V2V communication can use. Constraint C4 indicates the minimum signal-to-noise ratio of video semantic detection. Constraint C5 stipulates that the transmit power of PM cannot exceed its maximum transmit power. Constraint C6 indicates the processing capacity limit of PL. Constraint C7 stipulates that PM can only select PL or roadside base station as the destination for video semantic detection.

[0061] Step 4: The present invention determines that the resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system is an NP-hard problem. This judgment can be proved by simplifying it into a special case of a known NP-complete problem, thereby revealing that it does not have the possibility of being accurately solved in polynomial time. To this end, the present invention adopts a decomposition strategy to divide the resource allocation optimization problem into two independent stages: PM communication mode selection and wireless resource allocation scheme determination. In the first stage, PM communication mode selection depends on the quality of the PM's communication link, which is converted into a bipartite graph maximum weight matching problem, and then solved using the Hungarian algorithm. In the second stage, using the results of the first stage, the wireless resource allocation in the resource allocation optimization problem can be converted into a convex optimization problem.

[0062] Based on the resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system, PM communication mode selection and wireless resource allocation scheme design are carried out as follows:

[0063] Step 4.1: The purpose of selecting a communication mode for a PM is to ensure better communication link quality and reliability between the V2I communication link and the V2V communication link. A higher channel gain generally indicates better communication quality. Since the communication mode selection of each PM is independent of other vehicles, the sub-problem of PM communication mode selection in the resource allocation optimization problem between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system in the tth time slot is described as:

[0064]

[0065] Among them, h m,B represents the channel gain of the V2I communication link between the mth PM and the roadside base station in the tth time slot, h m,PL represents the channel gain of the V2V communication link between the mth PM and PL in the tth time slot.

[0066] Step 4.2: The equivalent problem obtained in step 4.1 can be transformed into the maximum weight matching (MWM) problem of bipartite graphs, such as Figure 3 As shown, the weighted bipartite graph has a vertex set and two non-intersecting subsets. Subset 1 represents the PM requesting assistance in perception, and subset 2 represents the task processing node. In the present invention, the link weight between the PM and the task processing node is represented by W, which is determined by the channel gain of the communication link. The larger the channel gain, the larger the link weight. In addition, due to the limited task processing capability of PL, it is assumed that at most N PMs can complete the perception task in PL.

[0067] Due to the MWM property of bipartite graphs, the Hungarian algorithm can be considered as an effective method to solve the above equivalence problem. The Hungarian algorithm iteratively creates a weighted bipartite graph and identifies and eliminates enhancement paths according to a given weight matrix to improve the current match. The enhancement path starts from an unmatched individual, alternately traverses an unmatched task and a matching individual (in reverse), and then returns to another unmatched individual. Adjusting the matching degree on the path can convert each PM and node pair from non-matching to matching, and vice versa, thereby improving the overall weight value. Once all improvement paths are exhausted, the current match becomes the best match, and the sum of the weights represents the maximum match. If there are other improvement paths, continue to improve until a perfect state is reached.

[0068] The sub-problem of PM communication mode selection is transformed into a maximum weight matching problem of a bipartite graph, and then the Hungarian algorithm is used to solve the maximum weight matching problem of the bipartite graph to obtain an initial result set; then the number of binary values ​​1 in the initial result set is counted. If the counted number is less than or equal to N, the initial result set is directly used as the final result set; if the counted number is greater than N, the weights corresponding to all binary values ​​1 in the initial result set are sorted from large to small, and the binary values ​​1 corresponding to the N+1th to the last weights in all the sorted weights are modified to binary values ​​0 to obtain the final result set; then each binary value in the final result set is used as the value of the binary variable for selecting the communication mode of each PM in the tth time slot, and l m The value of Complete the PM communication mode selection.

[0069] Step 4.3: After obtaining the results of the first stage, the resource allocation optimization problem exhibits a non-convex characteristic because it is affected by the time-varying channel state. However, within the short interval of any given time slot, the channel state can be reasonably regarded as fixed, thereby enabling the present invention to transform the original problem into a convex optimization problem for solution.

[0070] The resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system is transformed into a convex optimization problem, which can be described as:

[0071]

[0072] Step 4.4: Use the convex optimization algorithm to solve the convex optimization problem, obtain the bandwidth allocated to each PM in the tth time slot and the transmission power of each PM in the tth time slot, complete the design of the wireless resource allocation scheme for each PM in the tth time slot, and output the wireless resource allocation scheme for each PM in the tth time slot, as well as the objective function value of the wireless resource allocation scheme for all PMs in the tth time slot. Here, the convex optimization algorithm selects the convex optimization toolkit (convex programming, CVX) attached to Matlab R2016b.

[0073] In order to further verify the feasibility and effectiveness of the method of the present invention, a simulation experiment was carried out on the method of the present invention.

[0074] The simulation scenario parameter settings are listed in Table 1, and the channel models of the V2I and V2V communication links are shown in Table 2.

[0075] Table 1 Simulation scenario parameter settings

[0076]

[0077] Table 2 Channel models for V2I and V2V communication links

[0078]

[0079] Rice fading parameter

[0080]

[0081] Under the above simulation scenario parameter settings, the method of the present invention is compared with the existing resource allocation mechanism that only focuses on V2I.

[0082] Figure 4 A comparison chart of the video semantic detection accuracy of each PM when the vehicle queue travels 500 meters from the starting point is given, using the method of the present invention and the existing resource allocation mechanism that only focuses on V2I. Figure 4 It can be seen that, considering each PM in the vehicle queue as a unit, when the vehicle queue travels 500 meters from the starting point, the video semantic detection accuracy of each PM in the vehicle queue gradually decreases and tends to be stable as the PM-PL distance increases. This is because PMs closer to the PL tend to use more stable and lower-cost V2V communication links, while PMs farther from the PL tend to use V2I communication links; on the contrary, as the PM-PL distance increases, the performance of the existing resource allocation mechanism that only focuses on V2I gradually improves. This is because the rear PM is closer to the base station and can obtain better link gain. From Figure 4It can also be seen that in the method of the present invention, the performance of the rear PM in the vehicle queue is only slightly higher than the existing resource allocation mechanism that only focuses on V2I. This is because the selection of the V2I communication link becomes dominant.

[0083] Figure 5 The comparison chart of the average video semantic detection accuracy of all PMs in the vehicle queue when the vehicle queue is considered as a whole and the starting distance is set to 0 meters, 100 meters, 200 meters, 300 meters, 400 meters and 500 meters respectively using the method of the present invention and the existing resource allocation mechanism that only focuses on V2I is given. Figure 5 It can be seen that as the distance traveled by the vehicle queue increases, the V2I communication link continues to deteriorate, and the performance of the method of the present invention and the existing resource allocation mechanism that only focuses on V2I both decrease. However, the performance of the method of the present invention is significantly better than the existing resource allocation mechanism that only focuses on V2I.

[0084] In summary, the method of the present invention has obvious advantages over the existing resource allocation mechanism that only focuses on V2I, which fully demonstrates the feasibility and effectiveness of the method of the present invention.

Claims

1. A resource allocation method for semantic detection in a vehicle platoon mode, characterized in that The steps include: Step 1: Build a video semantic communication vehicle networking system model covering PL, PM, and roadside base stations equipped with EC servers: 1) Set up roadside base stations equipped with EC servers: Divide the highway into multiple consecutive sections, and set up a roadside base station equipped with an EC server in each section. The network signal of each roadside base station completely covers the section where it is located, and can communicate with any vehicle traveling in the section where it is located; 2) Set up a vehicle queue: The vehicle queue consists of 1 PL and M PMs, a total of M+1 vehicles, each PM is equipped with a high-definition camera; the time domain of the vehicle queue when it is running continuously is divided into multiple time slots, and all vehicles in the vehicle queue in a time slot are set to travel in the same interval; where M≥1; 3) Clarify the two tasks of the video semantic communication vehicle network system model, namely PM communication mode selection and wireless resource allocation scheme determination; In the video semantic communication vehicle network system model, each PM uses a high-definition camera to capture road information video in real time, and the PL and the roadside base station obtain their respective channel status data in real time. Then, the roadside base station selects the communication link with the relatively best channel status as the optimal communication link for each PM from the V2V communication link between the PM and the PL and the V2I communication link between the PM and the roadside base station based on its own channel status data and the channel status data of the PL, and considering the processing capacity limitation of the PL. Then, a node in the PL or roadside base station corresponding to the optimal communication link is determined as the receiving node. The PM encodes and compresses the road information video according to the wireless resource allocation scheme, and transmits the encoded road information video to the receiving node through the optimal communication link. The receiving node decodes and restores the received road information video, and performs semantic analysis on the decoded and restored road information video. The semantic analysis results are summarized to the PL to guide the overall driving behavior of the vehicle queue. Step 2: Construct a mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the m-th PM and PL perform V2V communication in the t-th time slot, which is described as: The mathematical relationship model between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system when the mth PM performs V2I communication with the roadside base station in the tth time slot is constructed, which is described as: Where, m = 1, 2, ..., M, f() is the function representation, It represents the accuracy of video semantic detection when the m-th PM and PL perform V2V communication in the t-th time slot. The function of It represents the accuracy of video semantic detection when the mth PM communicates with the roadside base station in the tth time slot. The function of represents the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot, represents the transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot, and They are calculated based on the bandwidth allocated to the mth PM and the transmission power of the mth PM in the tth time slot. The bandwidth and transmission power are the available wireless resources. α1, α2, α3, η1, η2 are all function parameters, and e is a natural constant. Step 3: Construct a resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system, which is described as: Among them, l m represents the binary variable for the communication mode selection of the mth PM in the tth time slot, l m =1 means that the V2V communication link between the mth PM and the PL is selected for the mth PM in the tth time slot, l m = 0 means that the V2I communication link between the mth PM and the roadside base station is selected for the mth PM in the tth time slot. m =1 when f(R m )for When l m = 0 when f(R m )for Rm=Bmlog2(1+γm), when l m =1 when Rm is When l m = 0 when R m for B m represents the bandwidth allocated to the mth PM in the tth time slot. m =1 when γ m for represents the received signal-to-noise ratio of PL when the mth PM and PL perform V2V communication in the tth time slot. m = 0 when γ m for represents the received signal-to-noise ratio of the roadside base station when the mth PM communicates with the roadside base station in the tth time slot, P m represents the transmission power of the mth PM in the tth time slot, represents the maximum transmission power of the mth PM, represents the maximum bandwidth that can be used by all PMs in V2V communication with PL, It represents the maximum bandwidth that can be used by all PMs for V2I communication with roadside base stations. represents the maximum time for the mth PM to transmit the compressed and coded road information video, γ th represents the preset signal-to-noise ratio threshold, N represents the number of nodes in PL, w1, w2, w3, and w4 are all weights, and their values ​​are all negative; Step 4: Based on the resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system, PM communication mode selection and wireless resource allocation scheme design are performed as follows: Step 4.1: Describe the sub-problem of PM communication mode selection in the resource allocation optimization problem between the video semantic detection accuracy and the amount of available wireless resources in the wireless communication system in the tth time slot as follows: Among them, h m,B represents the channel gain of the V2I communication link between the mth PM and the roadside base station in the tth time slot, h m,PL represents the channel gain of the V2V communication link between the mth PM and PL in the tth time slot; Step 4.2: Convert the subproblem about PM communication mode selection into the maximum weight matching problem of bipartite graph, and then use the Hungarian algorithm to solve the maximum weight matching problem of bipartite graph to obtain an initial result set; then count the number of binary values ​​1 in the initial result set. If the counted number is less than or equal to N, directly use the initial result set as the final result set; if the counted number is greater than N, sort the weights corresponding to all binary values ​​1 in the initial result set from large to small, and modify the binary values ​​1 corresponding to the N+1th to the last weight in all the sorted weights to binary value 0 to obtain the final result set; then use each binary value in the final result set as the value of the binary variable for communication mode selection of each PM in the tth time slot, and obtain l m The value of Complete the selection of PM communication mode; Step 4.3: The resource allocation optimization problem between the accuracy of video semantic detection in the tth time slot and the amount of available wireless resources in the wireless communication system is transformed into a convex optimization problem, which is described as: Step 4.4: Use the convex optimization algorithm to solve the convex optimization problem, obtain the bandwidth allocated to each PM in the tth time slot and the transmission power of each PM in the tth time slot, complete the design of the wireless resource allocation plan for each PM in the tth time slot, and output the wireless resource allocation plan for each PM in the tth time slot, as well as the objective function value of the wireless resource allocation plan for all PMs in the tth time slot.

2. The resource allocation method for semantic detection in vehicle platoon mode according to claim 1 is characterized in that In step 1, the duration of each time slot is 10 milliseconds.

3. The resource allocation method for semantic detection in vehicle platoon mode according to claim 1 is characterized in that The specific process of step 2 is: Step 2.1: Calculate the channel gain h of the V2V communication link between the mth PM and PL in the tth time slot m,PL , And calculate the channel gain h of the V2I communication link between the mth PM and the roadside base station in the tth time slot m,B , Among them, g m,PL represents the small-scale Rician fading with unit mean corresponding to the V2V communication link between the mth PM and PL in the tth time slot, g m,B represents the small-scale Rician fading with unit mean corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, β m,PL represents the shadow fading of the log-normal distribution corresponding to the V2V communication link between the mth PM and PL in the tth time slot, β m,B represents the shadow fading of the log-normal distribution corresponding to the V2I communication link between the mth PM and the roadside base station in the tth time slot, A represents the path loss constant, γ represents the path loss exponent, and L m,PL represents the distance between the mth PM and PL in the tth time slot, L m,B represents the distance between the mth PM and the roadside base station in the tth time slot; Step 2.2: Assume that LTE-V2X technology is used to implement V2V communication and V2I communication. The received signal-to-noise ratio of PL when the m-th PM and PL perform V2V communication in the t-th time slot is expressed as The received signal-to-noise ratio of the roadside base station when the mth PM performs V2I communication with the roadside base station in the tth time slot is expressed as Among them, P m represents the transmission power of the mth PM in the tth time slot, σ 2 represents the noise power density of additive white Gaussian noise at the receiver, where the receiver is a PL or a roadside base station, B m represents the bandwidth allocated to the mth PM in the tth time slot; Step 2.3: Express the transmission rate in the V2V communication link between the mth PM and PL in the tth time slot as The transmission rate in the V2I communication link between the mth PM and the roadside base station in the tth time slot is expressed as Step 2.4: The QP value when the mth PM and PL perform V2V communication in the tth time slot The relationship model is expressed as: Will The QP value when the mth PM communicates with the roadside base station in the tth time slot The relationship model is expressed as: Step 2.5: The video semantic detection accuracy of the case where the m-th PM and PL perform V2V communication in the t-th time slot is about The function is expressed as Combined with get The accuracy of video semantic detection in the case of V2I communication between the mth PM and the roadside base station in the tth time slot is about The function is expressed as Combined with get 4. The resource allocation method for semantic detection in vehicle platoon mode according to claim 1 is characterized in that In step 4.4, the convex optimization algorithm selects the convex optimization toolkit included in Matlab R2016b.

Images