Search Results

For intelligent vehicles, a robust perception system relies on training datasets with a large variety of scenes. The architecture of federated learning allows for efficient collaborative model iteration while ensuring privacy and security by leveraging data from multiple parties. However, the local data from different participants is often not independent and identically distributed, significantly affecting the training effectiveness of autonomous driving perception models in the context of federated learning. Unlike the well-studied issues of label distribution discrepancies in previous work, we focus on the challenges posed by scene heterogeneity in the context of federated learning for intelligent vehicles and the inadequacy of a single scene for training multi-task perception models. In this paper, we propose a federated learning-based perception model training system.

As a key technology of intelligent transportation system, vehicle type recognition plays an important role in ensuring traffic safety,optimizing traffic management and improving traffic efficiency, which provides strong support for the development of modern society and the intelligent construction of traffic system. Aiming at the problems of large number of parameters, low detection efficiency and poor real-time performance in existing vehicle type recognition algorithms, this paper proposes an improved vehicle type recognition algorithm based on YOLOv5. Firstly, the lightweight network model MobileNet-V3 is used to replace the backbone feature extraction network CSPDarknet53 of the YOLOv5 model. The parameter quantity and computational complexity of the model are greatly reduced by replacing the standard convolution with the depthwise separable convolution, and enabled the model to maintain higher accuracy while having faster reasoning speed.

As a key tool to maintain urban cleanliness and improve the road environment, road cleaning vehicles play an important role in improving the quality of life of residents. However, the traditional road cleaning vehicle requires the driver to monitor the situation of road garbage at all times and manually operate the cleaning process, resulting in an increase in the driver 's work intensity. To solve this problem, this paper proposes a road garbage recognition algorithm based on improved YOLOv5, which aims to reduce labor consumption and improve the efficiency of road cleaning. Firstly, the lightweight network MobileNet-V3 is used to replace the backbone feature extraction network of the YOLOv5 model. The number of parameters and computational complexity of the model are greatly reduced by replacing the standard convolution with the deep separable convolution, which enabled the model to have faster reasoning speed while maintaining higher accuracy.

Accurately predicting the future trajectories of surrounding traffic agents is important for ensuring the safety of autonomous vehicles. To address the scenario of frequent interactions among traffic agents in the highway merging area, this paper proposes a trajectory prediction method based on interactive graph attention mechanism. Our approach integrates an interactive graph model to capture the complex interactions among traffic agents as well as the interactions between these agents and the contextual map of the highway merging area. By leveraging this interactive graph model, we establish an agent-agent interactive graph and an agent-map interactive graph. Moreover, we employ Graph Attention Network (GAT) to extract spatial interactions among trajectories, enhancing our predictions. To capture temporal dependencies within trajectories, we employ a Transformer-based multi-head self-attention mechanism.

At present, most of the longitudinal car-following control algorithms based on model predictive control (MPC) do not consider the influence of the presence of the sloping road on the inter-vehicle distance, resulting in poor tracking capability under ramp conditions. In order to reduce the inter-vehicle distance error under ramp conditions and improve the tracking capability of longitudinal car-following control algorithm. The car-following control algorithm based on MPC considering uphill and downhill is proposed. This algorithm is based on the vehicle structure of fuel passenger cars, and adds a slope angle reconstruction module for implementing slope angle measurement and reducing the complexity of slope angle calculation based on the framework of conventional hierarchical control structure.

Preventing skidding is essential for studying the safety of driving in curves. However, the adhesion of the vehicle during the driving process on the wet and slippery road will be significantly reduced, resulting in lateral slippage due to the low adhesion coefficient of the road surface, the speed exceeding the turning critical, and the turning radius being too small when passing through the corner. Therefore, to reduce the incidence of traffic accidents of passenger cars driving in curves on rainy and snowy days and achieve the purpose of planning safe driving speed, this paper proposes a curve active safety system based on a deep learning algorithm and vehicle dynamics model. First,we a convolutional neural network (CNN) model is constructed to extract and judge the characteristics of snow and ice adhesion on roads.

Intelligent sweeper vehicle is gradually applied to human life, in which the accuracy of garbage identification and classification can improve cleaning efficiency and save labor cost. Although Deep Learning has made significant progress in computer vision and the application of semantic network segmentation can improve waste identification rate and classification accuracy. Due to the loss of some spatial information during the convolution process, coupled with the lack of specific datasets for garbage identification, the training of the network and the improvement of recognition and classification accuracy are affected. Based on the Unet algorithm, in this paper we adjust the number of input and output channels in the convolutional layer to improve the speed during the feature extraction part. In addition, manually generated datasets are used to greatly improve the robustness of the model.

Accurate surrounding vehicle motion prediction is critical for enabling safe, high quality autonomous driving decision-making and motion planning. Aiming at the problem that the current deep learning-based trajectory prediction methods are not accurate and effective for extracting the interaction between vehicles and the road environment information, we design a target vehicle intention-aware dual attention network (IDAN), which establishes a multi-task learning framework combining intention network and trajectory prediction network, imposing dual constraints. The intention network generates an intention encoding representing the driver’s intention information. It inputs it into the attention module of the trajectory prediction network to assist the trajectory prediction network to achieve better prediction accuracy.

Dangerous driving behavior will cause serious traffic accidents, which will not only threaten life and property but also cause traffic congestion and reduce road capacity. Speed detection is an important detection method to identify whether a driver is driving dangerously. Traditional speed detection methods need additional sensors, which will increase the cost of speed measurement. This paper proposes a vehicle speed estimation algorithm based on the imaginary projection plane (IPP). The IPP will be established according to the height, field angle, and vertical tilt angle of the camera and will be used to establish the mapping relationship between the world coordinates and image coordinates of the vehicle. By combining YOLOv4 and DeepSORT, the vehicle license plate is detected and tracked, and the center point of the vehicle license plate is taken as the feature point of vehicle speed estimation. The vehicle speed is estimated according to the IPP.

The sweeping vehicle has made a great contribution to the cleaning of urban roads. The traditional electric sweeping vehicle uses the main and auxiliary motors to drive the driving system and the operating system respectively. However, because the sweeper is in a low-speed working condition for a long time, and the drive motor must meet the demand for high power, there exist problems of low motor utilization and high cost. Aiming at this phenomenon, a dual-motor power coupling system based on planetary gears is proposed. First, analyze the driving mode of the dual-motor coupling power system according to the actual working scheme of the sweeper, and match the parameters of the motor based on this. Second, on the premise of meeting the power requirements, analyze and divide the working range of each drive mode based on the principle of minimum energy consumption, and then obtain the best drive mode switching control and speed and torque distribution strategy.

Road traffic accidents resulting from alcohol-impaired driving are increasing globally despite several measures, currently in place, to curb the trend. For this reason, recent research aims at integrating alcohol early-detection systems and driving simulator experiments to identify intoxicated drivers. However, driving simulator experiments on drunk driving have focused mostly on male participants than female drivers whose characteristics have scarcely been explored. Hence in this paper, vehicle dynamic control inputs on steering, braking, and acceleration performance of 75 licensed female drivers with an upshot of alcohol at four different blood alcohol concentration (BAC) levels (0%, 0.03%, 0.05%, and 0.08%) were investigated. The participants completed simulated driving in a fixed-based simulator experiment coupled with real-time ecological scenarios to extract discrete responses.

Heavy commercial vehicle drivers may frequently shift gears when they are running on long and downhill roads in mountainous area. In order to improve driving safety and fuel economy, it is necessary to predict the slope of the road ahead in real time and correct the driver's shift strategy in time. At present, the road slope estimation is mainly based on the real-time estimation of the road slope at the current position of the vehicle based on the vehicle driving information obtained by the sensors, but the road slope of the road section that the vehicle is about to reach has not been predicted. In this paper, based on the road slope information of the road section that the driver has driven through, combined with Geographic Information System (GIS) information and road design standards, the slope of the road section ahead is predicted.

The automatic emergency braking (AEB) system is an important part of automobile active safety, which can effectively reduce rear-end collision accidents and protect drivers' safety through active braking. AEB system has been included in many countries' new car assessment programme as the test content of active safety. In view of obviously deficiencies of the existing AEB control algorithm in avoiding longitudinal collision at high speed, it is proposed to an optimized model of the minimum safe distance for rear-end collision prevention on high-speed road in order to improve the accuracy of AEB system. Considering the influence of road adhesion coefficient and human comfort on the maximum braking deceleration, it is established to a more accurate and reasonable AEB system to avoid collision for expressway. The collision avoidance strategy is verified by simulation software.

The acquisition of vehicle dimensions in a vehicle’s moving process has a wide application in road monitoring, transportation, vehicle model recognition and non-contact overload recognition. At present, the detection of the vehicle dimensions mostly adopts the methods of human visual inspection and tool detection, which has a low detection efficiency and difficult to replicate on a large scale. Based on the image background subtraction method, this paper proposes a vehicle dimensions detection method, which can realize real-time detection of road vehicle dimensions. This method uses an adaptive Gaussian Mixture Model (GMM) to establish a background model based on the video stream. Initially, the moving target image is obtained by the background subtraction method, and then the edge detection under the Canny operator and Hough transform circle detection are performed on the image to obtain the pixel dimension of the vehicle's outline.

With the development of economy, the road cleaning faces great challenges because the road area keeps increasing and the road types tend to be diversified. Cleaning vehicle is widely used in road surface cleaning, but it is more and more difficult to meet the demand of road surface cleaning only through using a single road surface cleaning method. If the way of manual adjustment of cleaning parameters is adopted, the driver is required to have rich experience. At present, there is an urgent need for a cleaning vehicle that can autonomously adjust cleaning parameters according to the road surface. This study is based on road recognition technology. After the pavement category is reflected by the visual sensor feedback information and the pavement adhesion coefficient, the parameters of the cleaning vehicle are adjusted by the controller to adapt to different roads.

Discretionary lane changing is commonly seen in highway driving. Intelligent vehicles are expected to change lanes discretionarily for better driving experience and higher traffic efficiency. This study proposed to optimize the decision-making and trajectory-planning process so that intelligent vehicles made lane changes not only with driving safety taken into account, but also with the goal to improve driving comfort as well as to meet the driver’ s expectation. The mechanism of how various factors contribute to the driver’s intention to change lanes was studied by carrying out a series of driving simulation experiments, and a Lane-Changing Intention Generation (LCIG) model based on Bi-directional Long Short-Term Memory (Bi-LSTM) was proposed.

In ITS, image processing technology is applied to a wide variety of areas such as visual-based intelligent vehicle navigation, visual-based traffic monitoring and visual-based traffic management. In the recognition system of the vehicle body characteristics, most of the recognition is the license plate and the car emblem, etc. This paper proposes an image recognition application method for the vertical motion of the car while driving, mainly including vertical height detection and vertical displacement velocity acceleration recognition. The edge detection model of the image object is established by using the gray image to obtain the car motion segmentation image. At the same time, an image length and actual length coordinate conversion model is established, which can calculate an arbitrary actual length of the image object. In this paper, Yuejin Shangjun X500 van was selected as the test vehicle, and the video data was captured with a camera.

With the increase of car ownership and the complex and crowded parking environment, it is difficult for drivers to complete the parking operation quickly and accurately, which may cause traffic accidents such as vehicle collisions and road jams because of poor parking skills. The emergence of an automatic parking system can help drivers park safely and reduce the occurrence of safety accidents. In this paper, the neural network identifier on the control method of an adaptive integral derivative of a neural network is proposed for an automatic parallel parking system with front-wheel steering is studied by using MATLAB/Simulink environment, and the simulation is carried out. Firstly, according to vehicle parameters and obstacle avoidance constraints, the minimum parking space, and parking starting position are calculated. Meanwhile, the path planning of parallel parking spaces is carried out by quintic polynomial.

The ride performance and stability of the vehicle will decrease while the vehicle passing a deceleration bump with relatively high speed. If the speed is too low, the road efficiency and ride comfort will be affected. It is essential to identify the proper speed taking into account all the factors. In this paper, the dynamic model of the vehicle passing through the deceleration bump is established. Three kinds of indicators vibration weighted acceleration RMS, maximum vertical vibration acceleration and wheel load impact coefficient, are used to comprehensively evaluate the ride comfort and safety. The highway model, vehicle model, and common trapezoidal cross-sections bump models are set up in Carsim. Parameters such as vertical acceleration and tire force at different vehicle speeds are obtained. Then use the spline interpolation method to fit the data, and comprehensively consider the three indicators to get the best speed.

Deep Deterministic Policy Gradient (DDPG) is one of the Deep Reinforcement Learning algorithms. Because of the well perform in continuous motion control, DDPG algorithm is applied in the field of self-driving. Regarding the problems of the instability of DDPG algorithm during training and low training efficiency and slow convergence rate. An improved DDPG algorithm based on segmented experience replay is presented. On the basis of the DDPG algorithm, the segmented experience replay select the training experience by the importance according to the training progress to improve the training efficiency and stability of the training model. The algorithm was tested in an open source 3D car racing simulator called TORCS. The simulation results demonstrate the training stability is significantly improved compared with the DDPG algorithm and the DQN algorithm, and the average return is about 46% higher than the DDPG algorithm and about 55% higher than the DQN algorithm.