Vehicle cooperative perception data anti-interference transmission method for high-precision target detection

Abstract

The application discloses a vehicle cooperative perception data anti-interference transmission method for high-precision target detection and belongs to the fields of wireless communication and Internet of Vehicles. In the vehicle cooperative perception data transmission process for the high-precision target detection task, vehicles mine information such as a surrounding vehicle network topology, perception data acquisition time, a wireless channel state and interference intensity, adopt a reinforcement learning algorithm to dynamically optimize time slot selection and power allocation of Internet of Vehicles communication, efficiently share real-time perception data, reduce Internet of Vehicles cooperative perception data sharing delay and a packet loss rate, and further improve the precision of the vehicle cooperative target detection task.

Description

Technical Field

[0001] This invention belongs to the fields of wireless communication and vehicle networking, and in particular relates to an anti-interference transmission method for vehicle cooperative perception data for high-precision target detection. Background Technology

[0002] Vehicle-to-everything (V2X) cooperative perception for high-precision target detection tasks utilizes wireless communication to share complementary sensing feature information and leverages data fusion models to aggregate received multi-view feature data to enhance vehicle perception capabilities, such as trajectory prediction and obstacle detection beyond line-of-sight, effectively improving the safety of intelligent transportation. However, the time-varying channel states and interference signals in V2X can degrade the communication quality between vehicles, such as high packet loss rates and latency during sensing data sharing, thereby reducing vehicle perception accuracy and robustness, and even leading to serious traffic accidents. Therefore, an anti-interference and efficient transmission method for V2X cooperative perception data for high-precision target detection tasks is of great significance for ensuring perception accuracy and speed to improve the efficiency and safety of intelligent transportation.

[0003] Due to bandwidth limitations in vehicle-to-everything (V2X) communication, vehicles must balance collaborative perception performance with the amount of data transmitted when sharing perception data. The literature [Y.-C. Liu, et al.; “Who2com: Collaborative Perception via Learnable Handshake Communication,” in Proc. IEEE Int. Conf. Robot. Autom. (ICRA), Paris, France, 2020, pp. 6876–6883.] proposes a collaborative perception feature transmission scheme based on a three-way handshake. This scheme allows vehicles to select the most valuable collaborating vehicle based on request and response data packets, thereby improving perception performance and effectively reducing bandwidth requirements. The paper [Y.Hu, S.Fang, Z.Lei, Y.Zhong, S.Chen, “Where2comm: Communication-Efficient Collaborative Perception via Spatial Confidence Maps,” in Proc. Advances in Neural Inf. Process. Syst. (NeurIPS), New Orleans, LA, 2022.] proposes a collaborative perception scheme based on spatial confidence maps, enabling collaborative vehicles to share only sparse but critical spatial feature information, thereby improving the robustness of collaborative target detection while saving bandwidth. Furthermore, the communication latency generated by sharing perception data in vehicle-to-everything (V2X) communication may lead to a decrease in perception accuracy, posing potential risks in applications such as autonomous driving. CN202210816910.1 proposes a robust collaborative perception method addressing communication latency, employing a latency compensation module based on a long short-term neural network to predict vehicle positions and synchronize perception features with random latency, thereby reducing the impact of communication latency on perception accuracy.

[0004] Due to the limited computing resources of in-vehicle computing devices, the latency of computing services in vehicle-to-everything (V2X) applications may increase, thereby reducing the reliability of V2X services and creating potential risks in V2X security applications. The literature [M.Li, J.Gao, L.Zhao and X.Shen, “Deep Reinforcement Learning for Collaborative Edge Computing in Vehicular Networks,” in IEEE Trans. Cogn. Commun. Netw., vol.6, no.4, pp.1122-1135, Dec.2020.] proposes a task allocation and scheduling algorithm based on edge computing technology to allocate workloads and determine the execution order of tasks on edge servers. It also utilizes a deep deterministic policy gradient algorithm to optimize workload allocation and edge server selection in highly dynamic environments, thereby reducing computing service latency. The paper [Q. Luo, C. Li, TH Luan and W. Shi, “Collaborative Data Scheduling for Vehicular Edge Computing via Deep Reinforcement Learning,” in IEEE Internet Things J, vol. 7, no. 10, pp. 9637-9650, Oct. 2020.] proposes an integrated framework for communication, computing, caching, and collaborative computing based on in-vehicle edge computing. It employs deep reinforcement learning algorithms to optimize data scheduling strategies, thereby reducing data processing costs and processing latency of vehicle networking applications.

[0005] To reduce data transmission latency and improve communication efficiency in vehicle-to-everything (V2V) networks, the literature [H. Ye, GY Li and B.-HF Juang, “Deep Reinforcement Learning Based Resource Allocation for V2V Communications,” in IEEE Trans. Veh. Technol., vol. 68, no. 4, pp. 3163–3173, April. 2019.] uses reinforcement learning methods to optimize the transmission power and channel of V2V networks based on information such as interference level and channel gain, in order to meet the latency requirements of vehicle data sharing, such as reducing message sharing latency in cooperative sensing systems. The literature [Han Xu, Tian Daxin, Sheng, Zhengguo et al., Reliability-Aware Joint Optimization for Cooperative Vehicular Communication and Computing, in IEEE Trans. Intel. Transp. Syst., vol. 22, no. 8, pp. 5437-5446, 2021] uses a virtual queue model to optimize the data workload splitting point, so as to improve the coupling reliability of vehicle cooperative communication and computing and the task completion rate. Summary of the Invention

[0006] The purpose of this invention is to provide a method for anti-interference transmission of vehicle cooperative perception data for high-precision target detection, which uses reinforcement learning algorithms to dynamically optimize the time slot selection and power allocation during vehicle perception data transmission, reduces the latency and packet loss rate of perception data transmission under time-varying channel and signal interference conditions in vehicle networks, and improves the speed, accuracy and robustness of vehicle network cooperative target detection.

[0007] This invention includes the following steps:

[0008] Step 1: Let N be the number of vehicles participating in collaborative sensing in the vehicle-to-everything (V2X) network, denoted as... The number of available time slots and the transmit power order are denoted as M1 and M2, respectively, and the maximum transmit power is P. max Let the selectable time slots be X1 = {1, 2, ..., M1}, and the selectable transmit power be... Construct an M1×M2 dimensional action space vector and the observed state vector S;

[0009] Step 2: Create a table Q of size |X1|×|S| T and E Tand a table Q of size |X1|×|X2|×|S|. L and E L ;

[0010] Step 3: Initialize parameters, set initial delay Initial channel gain Packet loss rate The learning rate of the reinforcement learning algorithm is α = 0.4, and the discount factor is β = 0.3.

[0011] Step 4: In step k, acquire the point cloud data scanned by the LiDAR, and input it into a convolutional neural network to extract the perceptual features F of the point cloud data. (k) And record the acquisition time t of the perceived features. (k) ;

[0012] Step 5: Obtain the location of yourself and surrounding collaborative vehicles Channel states among neighboring vehicles participating in cooperative sensing are obtained through channel estimation. The current interference intensity is estimated using the received signal strength indication method.

[0013] Step 6: Quantize the channel state to order H to obtain the current estimated channel state. Based on the perception accuracy ρ fed back by the collaborative vehicles (k-1) Delay τ (k-1) Packet loss rate b (k-1) Calculate average sensing accuracy Average latency and average packet loss rate Construct a 2N+5 dimensional state vector

[0014] Step 7: s (k) Enter into form Q T and E T In the process, the corresponding |X1| dimension vector Q is obtained respectively. T (s (k) (x′1) and E T (s (k) ,x′1); where x′1 represents the state s (k) Any selectable time slot;

[0015] Step 8: Initialize the weight coefficients ξ T Time slots are selected based on the probability distribution of time slot selection.

[0016] Step 9: Construct a 2N+6 dimensional state vector

[0017] Step 10: Convert the 2N+6 dimensional state vector Enter form Q L and E L In the middle, the corresponding |X2| dimension vectors are obtained respectively. and Where x′2 represents the state Any selectable transmit power;

[0018] Step 11: Initialize the weight coefficients ξ L And based on the probability distribution of the transmission power selection Select transmit power

[0019] Step 12: In the time slot With transmission power Send perceptual features F (k) To support collaborative target detection tasks in vehicle-to-everything (V2X) networks;

[0020] Step 13: Receive the latency τ from the cooperating vehicle. (k) and perception accuracy ρ (k) Assess packet loss rate b (k) Calculate the average delay Average sensing accuracy and average packet loss rate

[0021] Step 14: Initialize weight coefficients c1 and c2, and calculate:

[0022]

[0023] Step 15: Initialize the weight coefficients c l,1 and c l,2 Risk level threshold μ l,1 and μ l,2 ,calculate:

[0024]

[0025] Where I(·) is an indicator function: it takes the value 1 when the variable is true, and 0 otherwise;

[0026] Step 16: Update Q T and Q L :

[0027]

[0028]

[0029] Step 17: Initialize the weighting coefficients β 0 and β 1 ,β 2,...,β L Update E T and E L :

[0030]

[0031] Step 18: Repeat steps 2 through 17 until the condition is met. And |Q T (s (k +1) ,x1 (k+1) )-Q T (s (k) ,x1 (k) If | < 0.01, the algorithm has converged.

[0032] Compared with the prior art, the present invention has the following outstanding advantages:

[0033] The present invention uses a reinforcement learning algorithm to dynamically optimize the time slot selection and power allocation of vehicle-to-everything (V2X) communication based on information such as the surrounding vehicle topology network, the sensing data acquisition time, the estimated wireless channel state and interference intensity, as well as the sensing accuracy, latency and packet loss rate fed back by the cooperating vehicles. This reduces the latency and packet loss rate of V2X collaborative sensing data sharing, thereby improving the accuracy of vehicle-to-everything (V2X) collaborative target detection tasks.

[0034] This invention proposes an anti-interference transmission method for vehicle cooperative sensing data for high-precision target detection, supporting vehicle-to-everything (V2X) cooperative target detection tasks based on LiDAR point cloud data. The method involves wireless communication, computer science, and V2X fields. It mines information such as the surrounding vehicle network topology, sensing data acquisition time, wireless channel status, and interference intensity. A reinforcement learning algorithm is used to dynamically optimize the time slot selection and power allocation for sensing data transmission to efficiently share real-time sensing data, reducing data transmission latency and packet loss rate, and improving the accuracy and sensing speed of the cooperative target detection task. Attached Figure Description

[0035] Figure 1 This refers to the data sharing delay in this embodiment of the invention.

[0036] Figure 2 This refers to the packet loss rate for data sharing in this embodiment of the invention.

[0037] Figure 3 This refers to the target detection accuracy in this embodiment of the invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments will be used in conjunction with the accompanying drawings to further illustrate the invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0039] The embodiments of the present invention include the following steps:

[0040] Step 1: Let N = 5 be the number of vehicles participating in collaborative perception in the vehicle-to-everything (V2X) network, expressed as... The number of available time slots and the order of transmit power are M1=10 and M2=10, respectively, and the maximum transmit power is P. max =100, let the selectable time slots be X1={1,2,...,10}, the selectable transmit power be X2={10,20,30,...,100}, and construct a 100-dimensional action space vector X={[1,10],[1,20],...,[10,100]} and a 4050-dimensional state space vector S.

[0041] Step 2: Create a table Q with a size of 40500. T and E T and a table Q of size 405000. L and E L .

[0042] Step 3: Initialize parameters, set initial delay Initial channel gain Packet loss rate The learning rate α = 0.4 and the discount factor β = 0.3 for the reinforcement learning algorithm.

[0043] Step 4: In the k-th time slot, acquire the point cloud data scanned by the LiDAR, and input it into a convolutional neural network to extract the perceptual features F of the point cloud data. (k) And record the acquisition time t of the perceived features. (k) .

[0044] Step 5: Obtain the location of yourself and surrounding collaborative vehicles Channel states among neighboring vehicles participating in cooperative sensing are obtained through channel estimation. The current interference intensity is estimated using the received signal strength indication method.

[0045] Step 6: Quantize the channel state to H=3 to obtain the current estimated channel state. And based on the perception accuracy ρ fed back by the collaborative vehicles (k-1) Delay τ (k-1) Packet loss rate b (k-1) Calculate average sensing accuracy Average latency and average packet loss rate Construct a 15-dimensional state vector

[0046] Step 7: s (k) Enter into form QT and E T In the middle, the corresponding 10-dimensional vector Q is obtained respectively. T (s (k) (x′1) and E T (s (k) (x′1). Where x′1 represents the state s. (k) Any selectable time slot.

[0047] Step 8: Initialize the weight coefficients ξ T =10, and based on the probability distribution of time slot selection. Select time slot

[0048] Step 9: Construct a 16-dimensional state vector

[0049] Step 10: Enter form Q L and E L In the middle, the corresponding 10-dimensional vectors are obtained respectively. and Where x′2 represents the state Any selectable transmit power.

[0050] Step 11: Initialize the weight coefficients ξ L =0.5, and based on the probability distribution selected according to the transmission power. Select transmit power

[0051] Step 12: In the time slot With transmission power Send perceptual features F (k) To support collaborative target detection tasks in vehicle-to-everything (V2X) networks.

[0052] Step 13: Receive the latency τ from the cooperating vehicle. (k) and perception accuracy ρ (k) Assess packet loss rate b (k) And further calculate the average delay. Average sensing accuracy and average packet loss rate

[0053] Step 14: Initialize weight coefficients c1 = 0.02, c2 = 4, and calculate:

[0054]

[0055] Step 15: Initialize the weight coefficients c l,1 =0.5,c l,2 =0.5, risk level threshold μ l,1=50 and μ l,2 =0.1, calculate:

[0056]

[0057] Where I(·) is an indicator function: it takes the value 1 when the variable is true, and 0 otherwise.

[0058] Step 16: Update Q T and Q L :

[0059]

[0060]

[0061] Step 17: Initialize the weighting coefficients β 0 =0.2 and β 1 =β 2 =...=β L =0.1, update E T and E L :

[0062]

[0063] Step 18: Repeat steps 2-17 until the condition is met. and That is, the algorithm converges.

[0064] Figure 1 The data sharing latency of an embodiment of the present invention is given. Figure 2 The data sharing packet loss rate of an embodiment of the present invention is given. Figure 3 The target detection accuracy of embodiments of the present invention is given. Figures 1-3 As can be seen, the embodiments of the present invention can improve the detection accuracy of the vehicle-to-everything (V2X) collaborative target detection task and reduce the data sharing latency and packet loss rate in the vehicle collaborative perception process.

[0065] Unlike existing collaborative sensing schemes, this invention proposes an anti-interference transmission method for vehicle collaborative sensing data for high-precision target detection tasks. It mines information such as the surrounding vehicle topology network, sensing data acquisition time, estimated wireless channel state, and interference intensity, as well as the sensing accuracy, latency, and packet loss rate fed back by collaborative vehicles. A reinforcement learning algorithm is used to dynamically optimize the time slot selection and power allocation of vehicle-to-everything (V2X) communication, reducing the latency and packet loss rate of V2X collaborative sensing data sharing, thereby improving the accuracy of vehicle collaborative target detection tasks.

[0066] The above embodiments are merely preferred embodiments of the present invention and should not be considered as limiting the scope of the present invention. All equivalent variations and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for anti-interference transmission of vehicle cooperative perception data for high-precision target detection, characterized in that... Includes the following steps: Step 1: Let the number of vehicles participating in collaborative perception in the vehicle-to-everything (V2X) network be... , represents The number of available time slots and the transmit power order are denoted as follows: and Maximum transmission power is , record optional time slots Selectable transmit power , build 3D action space vector and the observed state vector S; Step 2: Create a size of table and and size table and ; Step 3: Initialize parameters, set initial delay Initial channel gain Packet loss rate The learning rate of reinforcement learning algorithms Discount factor ; Step 4: In the The first step is to acquire point cloud data scanned by the LiDAR, and then input the data into a convolutional neural network to extract the perceptual features of the point cloud data. And record the acquisition time of the perceived features. ; Step 5: Obtain the location of yourself and surrounding collaborative vehicles Channel states among neighboring vehicles participating in cooperative sensing are obtained through channel estimation. The current interference strength is estimated by the received signal strength indication method. ; Step 6: Quantize the channel state as The order is used to obtain the current estimated channel state. Based on the perception accuracy fed back by the collaborative vehicles Delay Packet loss rate Calculate average sensing accuracy Average latency and average packet loss rate Construct a 2N+5 dimensional state vector ; Step 7: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Enter into the table and In the middle, the corresponding dimensional vector and ;in, Indicates the state Any selectable time slot; Step 8: Initialize weight coefficients Time slots are selected based on the probability distribution of time slot selection. ; Step 9: Construct a 2N+6 dimensional state vector ; Step 10: Convert the 2N+6 dimensional state vector Enter form and In the middle, the corresponding dimensional vector and ;in, Indicates the state Any selectable transmit power; Step 11: Initialize weight coefficients And based on the probability distribution of the transmission power selection Select transmit power ; Step 12: In the time slot With transmission power Send sensing features To support collaborative target detection tasks in vehicle-to-everything (V2X) networks; Step 13: Receive latency feedback from collaborating vehicles and perception accuracy Assess packet loss rate Calculate the average delay Average sensing accuracy and average packet loss rate ; Step 14: Initialize weight coefficients , ,calculate: Step 15: Initialize weight coefficients and Risk level threshold and ,calculate: in, It is an indicator function: it takes the value 1 when the variable is true, and 0 otherwise; Step 16: Update and : Step 17: Initialize weight coefficients and ,renew and : Step 18: Repeat steps 2 through 17 until the condition is met. and This means the algorithm has converged.

Images