To promote the application of core technologies in the autonomous operation system of intelligent and connected vehicles, this article analyzed the development trends, technical architectures and key technologies of future automotive operation systems from the perspectives of vehicle electronic and electrical architectures. At the same time, based on the application needs of vehicle manufacturers, the paper proposed implementation suggestions to accelerate the production and installation of domestic operation systems and ecological construction, in order to help accelerate the industrialization and development of autonomous operation systems for intelligent and connected vehicles.

For the problem that intelligent vehicles lack the driver’s subjective driving operation cognition and the risk cognition of traffic participants, this paper proposed a trajectory planning algorithm based on improved Driver Risk Field (DRF). Firstly, the coordinate transformation from Cartesian coordinate system to Frenet coordinate system was carried out for the position of the vehicle, which visually represented the position information between the vehicle and the road; Secondly, a driver risk field model integrating vehicle stability risk was constructed to perceive the risk of traffic participants from the driver’s perspective; Finally, considering the comfort needs of drivers and passengers, obstacle avoidance operations were performed on the perceived risk potential energy high points based on quintic and quartic polynomial trajectories. The simulation and hardware-in-the-loop test results show that the planned trajectory meets both the obstacle avoidance function and the acceleration constraints, ensuring the safety and comfort of driving.

To make autonomous lane change for intelligent vehicles in the situations with no lane or unclear lane, this paper proposed a lane change trajectory planning method based on virtual lane lines. Firstly, a virtual lane was constructed to divide the road into lanes parallel to the vehicle orientation, the best lane was determined according to the minimum heading angle of collision-free lane change. Secondly, the lane change end state was determined according to the average vehicle speed of the current lane and the target lane. The lane change trajectory was generated using the quintic polynomial combined with vehicle kinematics and dynamics. The path planning algorithm proposed was added to the intelligent planning decision-making module of the intelligent vehicle for simulation and hardware-in-loop test. The results show that the method realizes the function of autonomous lane changing in the situations without lane and improve the safety and driving efficiency of the vehicle in complex traffic environment.

In order to improve the environmental perceive ability and better predict the behaviors of dynamic vehicle targets, a YOLOX model, as the front-end detector, combined with the optimized DeepSort method was proposed to develop the research of dynamic vehicle Multi-Object Tracking. In the process of features matching for dynamic vehicles, the change information of vehicles in the degree of light and shade was extracted to improve the object matching accuracy by adding Haarlike features. Furthermore, based on the application of DeepSort RE-identification network, the improved ResNet13 model was adopted as the backbone network for feature extraction, and SENet was introduced to adjust the feature weight of different channel dimensions. The results show that, compared to the conventional DeepSort method, by using the real on-road video data, the Multi-Object Tracking Accuracy (MOTA) and IDF1 indicator of the proposed algorithm increased by 1.4 percentage points and 7.7 percentage points respectively.

For the problem that the autonomous navigation exploration algorithm is easy to fall into the local area, this paper proposed an exploration algorithm combining sampling and deep reinforcement learning. First, the Long-Short-Term Memory (LSTM) network was used locally to obtain the historical pose information of the unmanned vehicle to avoid repeated exploration of the explored area; secondly, the optimal action of the deep reinforcement learning strategy was used to output using deep reinforcement learning and the reward function was designed to encourage the unmanned vehicle to fully explore the unknown area; Finally, the horizontal movement factor of the unmanned vehicle was considered to generate a global exploration path conforming to its current attitude by solving the Asymmetric Travel Salesman Problem (ATSP). In the 2 000 s mine tunnel simulation environment, compared with the Technologies for Autonomous Robot Exploration (TARE) algorithm, the proposed algorithm increased the exploration area by 346.3 m² and reduced the total driving distance by 209.4 m; in the real scene test, the exploration algorithm completed the exploration of the underground garage with an area of 3 444.3 m² and returned to the starting point in 1 014 s and built the environment map.

In order to improve the real-time and accuracy of parking slot detection in automatic parking, this paper proposed a lightweight parking-slot detection algorithm based on collaborative attention and graph neural network. Firstly, This algorithm used a lightweight network structure and the improved MobileNetV3 as the feature extraction network, obtained the location information and feature information of the parking-slot marker points through depthwise separable convolution, combined them to obtain the fused features of the marker points, then constructed a graph network structure to enhance the internal relationship of the parking-slot marker points, and combined the cooperative attention mechanism to integrate multiple attention. Finally, the algorithm was tested on the public parking-slot dataset PS2.0. The results indicate that the detection accuracy is better than the current mainstream algorithm, the average reasoning speed of each frame of image can reach 10.1 ms, with good accuracy and real-time performance.

For the problem of communication delay caused by the interference of the communication channel in the process of vehicle-road cooperative V2X communication, this paper proposed a V2X delay attack protection technology based on game theory. With the signal-to-noise ratio as the measurement index of communication quality, this paper firstly studied the function relationship between the transmission power of the attacking node and the signal-to-noise ratio to obtain the function relationship between the transmission power of the legitimate node and the attacking node, then this functional relationship was brought into the relationship between transmission power of legitimate node and the signal-to-noise ratio. This function also considered the probability of signal transmission, the probability of being detected and the transmission interference of other nodes to obtain the target function. Finally, the proposed protection technology is verified by simulation and test. The results show that by adjusting the transmission power of the nodes, it can effectively resist the attacker’s interference on the V2X communication channel and reduce the communication delay.

To ensure the security and privacy of sensitive data in Internet of Vehicle (IoV) environments, this paper proposed a distributed differential privacy data protection scheme combining federated learning and reinforced learning mechanisms. In this scheme, a federated learning architecture was applied to keep data on vehicle nodes or edge devices for learning, enabling data privacy protection, reducing data transmission costs through distributed storage. The Laplace mechanism was employed to achieve differential privacy, the Layer-wise Relevance Propagation (LRP) was used to manage data perturbation, ensuring the privacy and efficiency of model parameter transmissions. Experimental results show that the proposed scheme can achieve approximately 80% global accuracy within 10 rounds of communication, with a maximum of 98%, can complete model aggregation within less communication rounds, achieving a good balance between privacy protection and global data accuracy, and accurately detecting the injection of false noise through the reinforced learning strategy, promoting the intelligence and security levels of IoV.