Search

Adversarial patches can disrupt computer vision systems, seriously threatening people's lives and property security. Existing defense methods seldom consider the generalization for defending against different patches and the compatibility with various models. Furthermore, the severe security situation necessitates the combination of data defense and model defense to build a comprehensive defense system. To address these issues, we propose a defense method named PatchBreaker, which consists of three components. In data defense, PatchBreaker uses the Semantic-Cutter trained by annotated patch images to cut patches and output incomplete images. Next, the Image-Inpainter trained by clean-incomplete image pairs is used to inpaint these incomplete images and output inpainted images. In model defense, the Adversaril-Classifier will be trained by joint adversarial training with clean images and patch images. Finally, PatchBreaker inputs inpainted images into Adversarial-Classifier to output correct results. Comparative experiments show that PatchBreaker outperforms other comparative defense methods in most cases, which indicates the excellent patch generalization and model compatibility of PatchBreaker. Meanwhile, ablation studies show the effectiveness of combining data defense and model defense. Additionally, PatchBreaker has minimal impact on the clean accuracy (about 1 % ). The background, mechanisms and defense effectiveness of PatchBreaker. Intelligent systems are easily disrupted by adversarial patches. Therefore, we propose the PatchBreaker, which consists of data defense and model defense. In data defense, the Semantic-Cutter cuts patches by BiseNetV2 model and output incomplete images, then the Image-Inpainter inpaints incomplete images by a bilateral image inpainting model. In model defense, we utilize clean and patch images to train the Adversarial-Classifier for classifying inpainted images to output correct results. After defense, the green high-light regions of integrated gradients attribution are more regular, which indicates the effectiveness of PatchBreaker [ABSTRACT FROM AUTHOR]

Multisensor (MS) data fusion is important for improving the stability of vehicle environmental perception systems. MS joint calibration is a prerequisite for the fusion of multimodality sensors. Traditional calibration methods based on calibration boards require the manual extraction of many features and manual registration, resulting in a cumbersome calibration process and significant errors. A joint calibration algorithm for a Light Laser Detection and Ranging (LiDAR) and camera is proposed based on deep learning without the need for other special calibration objects. A network model constructed based on deep learning can automatically capture object features in the environment and complete the calibration by matching and calculating object features. A mathematical model was constructed for joint LiDAR-camera calibration, and the process of sensor joint calibration was analyzed in detail. By constructing a deep-learning-based network model to determine the parameters of the rotation matrix and translation matrix, the relative spatial positions of the two sensors were determined to complete the joint calibration. The network model consists of three parts: a feature extraction module, a feature-matching module, and a feature aggregation module. The feature extraction module extracts the image features of color and depth images, the feature-matching module calculates the correlation between the two, and the feature aggregation module determines the calibration matrix parameters. The proposed algorithm was validated and tested on the KITTI-odometry dataset and compared with other advanced algorithms. The experimental results show that the average translation error of the calibration algorithm is 0.26 cm, and the average rotation error is 0.02°. The calibration error is lower than those of other advanced algorithms. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Chongqing University of Technology (Natural Science) is the property of Chongqing University of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

A new energy vehicle cab control platform based on human-computer interaction is proposed from control device design and control signal acquisition design. A driving mode conversion algorithm based on the threshold is combined with the existing intelligent vehicle driving mode conversion. The MATLAB and Python hybrid program is used to realize the simulation experiment of the driving control algorithm of new energy vehicles. The system can conduct interactive simulations of various typical driving scenarios. Data acquisition and dynamic display functions can present the operating status of new energy vehicles most directly and efficiently. The system can integrate and intelligently process vehicle driving data, road environment data and external data from multiple dimensions of the "vehicle-road-network". The experimental results show that the system enhances the user experience of new energy vehicles and improves the efficiency of human-computer interaction. [ABSTRACT FROM AUTHOR]

Objective The existing target speed planning algorithm for the automatic driving of urban rail transit train cannot handle in real time the temporary change of the line speed limit in case of emergency due to the large computational load and long computation time. With regard to the problem, a real-time target speed planning algorithm based on global optimization is proposed to generate the speed planning curve in real time. Method Based on the conditions of train current position, current speed and speed limit of the forward line, the speed planning curve is firstly generated in the shortest time through point by point calculation. Then the traction and braking levels are being adjusted to keep the train running at uniform acceleration or deceleration, optimizing the comfort index of the train running. Next, the cruising speed in the maximum speed limit section is adjusted to reduce unnecessary traction and braking time, optimizing the energy consumption index of the train running. Finally, the train speed planning curve is output. Result & Conclusion The simulation results show that the real-time speed planning curve generated by the proposed algorithm satisfies the basic constraints of safety, punctuality and accurate parking. Compared with the traditional algorithm, it improves the comfort level and reduces the energy consumption during train operation. Meanwhile, the proposed algorithm can effectively handle the temporary change of the line speed limit in emergency and optimize several operation indicators.

As an important part of the unmanned driving system, the detection and recognition of traffic sign need to have the characteristics of excellent recognition accuracy, fast execution speed and easy deployment. Researchers have applied the techniques of machine learning, deep learning and image processing to traffic sign recognition successfully. Considering the hardware conditions of the terminal equipment in the unmanned driving system, in this research work, the goal was to achieve a convolutional neural network (CNN) architecture that is lightweight and easily implemented for an embedded application and with excellent recognition accuracy and execution speed. As a classical CNN architecture, LeNet-5 network model was chosen to be improved, including image preprocessing, improving spatial pool convolutional neural network, optimizing neurons, optimizing activation function, etc. The test experiment of the improved network architecture was carried out on German Traffic Sign Recognition Benchmark (GTSRB) database. The experimental results show that the improved network architecture can obtain higher recognition accuracy in a short interference time, and the algorithm loss is significantly reduced with the progress of training. At the same time, compared with other lightweight network models, this network architecture gives a good recognition result, with a recognition accuracy of 97.53%. The network structure is simple, the algorithm complexity is low, and it is suitable for all kinds of terminal equipment, which can have a wider application in unmanned driving system. [ABSTRACT FROM AUTHOR]

Copyright of Urban Mass Transit is the property of Urban Mass Transit Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Journal of Japan Society for Fuzzy Theory & Intelligent Informatics is the property of Japan Society for Fuzzy Theory & Intelligent Informatics and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Journal of Zhengzhou University: Engineering Science is the property of Editorial Office of Journal of Zhengzhou University: Engineering Science and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Automatic driving vehicles need to have the ability to predict the intentions of changing lanes to ensure driving safety in mixed traffic. In order to predict the intention as early as possible, a prediction model based on M-LSTM(morgrifier long short-term memory)network was proposed. First, the S-G (Savitzky-Golay) filter was used to filter the noise reduction of the natural driving data set NGSIM (next generation simulation), and the track sequence of different lengths of time was marked by changing lane to the left, right, and driving straightly, the input model of vehicle motion information and environmental information was selected. Finally, the softmax function was used to classify the intention. The result shows that the prediction accuracy of the model is higher than SVM (support vector machine) and LSTM under different prediction times, and the closer to the lane-changing point, the higher the prediction accuracy. At 1. 0 s and 2. 5 s, the prediction accuracy is 93. 83% and 81. 30% respectively. The proposed model has pleasurable accuracy and predictability. It can provide technical support for automatic driving vehicles to identify lane-changing intention as early as possible. [ABSTRACT FROM AUTHOR]

The detection of out-of-distribution (OoD) samples in semantic segmentation is crucial for autonomous driving, as deep learning models are typically trained under the assumption of a closed environment, whereas the real world presents an open and diverse set of scenarios. Existing methods employ uncertainty estimation, image reconstruction, and other techniques for OoD sample detection. We have observed that different classes may exhibit connections and associations in varying contexts. For example, objects encountered by autonomous vehicles differ in rural road scenes compared to urban environments, and the likelihood of encountering novel objects varies. This aspect is missing in current anomaly detection methods and is vital for OoD sample detection. Existing approaches solely consider the relative significance of each prediction class, overlooking the inter-object correlation. Although prediction scores (e.g., max logits) obtained from the segmentation network are applicable for OoD sample detection, the same problem persists, particularly for OoD objects. To address this issue, we propose the utilization of the Mahalanobis distance of max logits to evaluate the final predicted score. By calculating the Mahalanobis distance, the paper aims to uncover correlations between different classes, thus enhancing the effectiveness of OoD detection. To this end, we also extend the state-of-the-art segmentation model, DeepLabV3+, to enable OoD sample detection in this paper. Specifically, this paper proposes a novel backbone network, SOD-ResNet101, for extracting contextual and multi-scale semantic information, leveraging the class correlation feature of the Mahalanobis distance to enhance the detection performance of out-of-distribution objects. Notably, our approach eliminates the need for external datasets or separate network training, making it highly applicable to existing pretraining segmentation models. [ABSTRACT FROM AUTHOR]

In recent years, the performance of image-based object detection algorithms has improved significantly, especially in the field of autonomous driving. It is well known that camera sensors are susceptible to adverse weather conditions, which can significantly affect their performance. In contrast, millimeter wave radar is robust to such weather conditions. As a result, the fusion of millimeter-wave radar and camera sensor has gained considerable attention as a promising approach for object detection. However, existing methods hardly take into account the correlation between the two modalities, leading to detection results that are vulnerable to radar noise, visual blur, and other confounding factors. To address this challenge, we propose an interactive guidance network that leverages a cross-modal attention mechanism, enabling radar and camera sensors to mutually guide each other and learn the underlying correlation between the two modalities. Our approach aims to achieve complementary fusion of features while effectively utilizing information from both radar and camera sensors to enhance detection results. Moreover, a bi-directional fusion Feature Pyramid Network (FPN) structure is introduced, which generates feature maps with enhanced semantic and texture information. To assess the effectiveness of our proposed method, we conducted experiments on the NuScenes dataset. The results demonstrate that our approach outperforms existing state-of-the-art methods in terms of object detection accuracy. [ABSTRACT FROM AUTHOR]

Lane detection is one of the core technologies of assisted driving and automatic driving. To further enhance the ability of lane feature extraction, a lane line recognition algorithm based on deep learning and recurrent multi-feature information aggregator was proposed. Given the problem of model operation speed, the algorithm took the lane detection problem as a classification problem based on row selection cells. Given the lane information aggregation problem in the image, a novel RMFA(recurrent multi-feature information aggregator) method was proposed. The ResNet (residual neural network) was combined with the method to propose a lane line recognition network ResNet-RMFA fused with contextual and multi-channel information. The performance of the network model was tested on the open datasets of Tusimple and CULane. The experimental results show that the reasoning time of a single frame image of the model can reach 4. 8 ms, the accuracy on the Tusimple dataset is 96. 07%, and the F1 score on the CULane dataset (IoU =0. 5) is 69. 3%, achieving a good balance between speed and accuracy. [ABSTRACT FROM AUTHOR]

Automatic driving technology has high accuracy and real-time requirements for pedestrian identification and localization. Pedestrian detection is a basic and necessary function in vision-based pedestrian detection systems and collision warning, which can effectively avoid traffic accidents and improve road driving safety to a certain extent. In this paper, a lightweight solution based on the FCOS-DEFPN model is proposed for real-time pedestrian detection. Based on the FCOS model, this paper proposes the FCOS-DEFPN model, which achieves the lightweight of the network by replacing the ResNet50 backbone network with the MobilenetV3 network and using the depth separable convolution instead of the ordinary convolution for parameter compression. While maintaining the detection accuracy, this paper introduces data enhancement methods such as Random Erasing and Morsia to simulate pedestrian occlusion and small target scenarios to improve the robustness of the model. For the pedestrian occlusion scenario, this paper introduces a lightweight attention network ECA, which helps to extract pedestrian features better. For small-target multi-scale pedestrians, the DEFPN feature pyramid network is proposed, which acquires feature information at multiple scales by attentional fusion of feature layers at different scales from top-down, bottom-up, and front-back. The experimental results show that the proposed model is enhanced in terms of detection accuracy for occluded and small-target pedestrians, and satisfies real-time pedestrian detection under the premise of robustness in complex scenes.

Previous research on automatic control of high-speed trains in speed limit sections is insufficient. This article proposes a new offline reinforcement learning strategy for automatic tracking of autonomous trains. Firstly, the operating speed and deceleration starting point were determined for different speed limit scenarios. Then, a tracking controller based on the improved offline conservative Q-learning (CQL) algorithm was designed to avoid frequent interaction between the train and the environment. Selected an appropriate policy to implement the CQL algorithm. The data samples were reclassified to increase sample concentration. The value and strategy network structure was redesigned. The state space and action space of tracking trains were limited, and the dimension of state space was increased. A multi-objective reward function was designed to distinguish the tracking process of trains in different sections. The simulation results show that the proposed high-speed railway tracking interval automatic control algorithm is superior to traditional online reinforcement learning methods in terms of safety, comfort, and convergence efficiency.

Multisensor (MS) data fusion is important for improving the stability of vehicle environmental perception systems. MS joint calibration is a prerequisite for the fusion of multimodality sensors. Traditional calibration methods based on calibration boards require the manual extraction of many features and manual registration, resulting in a cumbersome calibration process and significant errors. A joint calibration algorithm for a Light Laser Detection and Ranging (LiDAR) and camera is proposed based on deep learning without the need for other special calibration objects. A network model constructed based on deep learning can automatically capture object features in the environment and complete the calibration by matching and calculating object features. A mathematical model was constructed for joint LiDAR-camera calibration, and the process of sensor joint calibration was analyzed in detail. By constructing a deep-learning-based network model to determine the parameters of the rotation matrix and translation matrix, the relative spatial positions of the two sensors were determined to complete the joint calibration. The network model consists of three parts: a feature extraction module, a feature-matching module, and a feature aggregation module. The feature extraction module extracts the image features of color and depth images, the feature-matching module calculates the correlation between the two, and the feature aggregation module determines the calibration matrix parameters. The proposed algorithm was validated and tested on the KITTI-odometry dataset and compared with other advanced algorithms. The experimental results show that the average translation error of the calibration algorithm is 0.26 cm, and the average rotation error is 0.02°. The calibration error is lower than those of other advanced algorithms.

Point cloud analysis is an important part of 3D geometric processing. It has been widely used in many fields, such as automatic driving and robots. Although great progress has been made in recent years, there are still some unresolved problems. For example, current methods devote employing MLP to extract local features after search k neighbor points, they cannot effectively model the dependency relationship between the anchor point and k neighboring points. In addition, the prevailing models may not exploit the inherent structural similarities present in the global scope. To solve these issues, we propose a feature extraction model named DetailPoint to get detailed local information and long-range global dependency of point clouds. DetailPoint possess three units: the shallow local learning unit, the deep local learning unit and the deep global learning unit. We first use the SLL to extract shallow local features, and then use the DLL to learn deep local features. In these two units, we design a dual-path extraction method to acquire detail local features with dependencies. Finally, the DGL unit is employed to improve the generalization ability of local features and establish global interaction. These three units are connected in series to form our DetailPoint. We evaluated the performance of our model on four datasets, ScanObjectNN and ModelNet40 for shape classification, the ShapeNet dataset for part segmentation, and the S3DIS dataset for sementatic segmentations. The experimental results demonstrate that DetailPoint is capable of expressing point clouds more effectively, resulting in superior performance compared to existing methods. [ABSTRACT FROM AUTHOR]

Copyright of Automotive Engineer (1674-6546) is the property of Auto Engineering Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The 3D perception system functions as one of the core components for the safe operation of autonomous-rail rapid tram. Addressing the limitations of LiDAR systems used for autonomous-rail rapid tram in complex urban environments, such as challenges in perceiving distant objects, insensitivity to color, and potential failure, this paper proposes a pure-vision monocular 3D perception system for autonomous-rail rapid tram. Comprising data preprocessing, model training and model deployment, this vision-based scheme with a full coverage from data collection to on-board deployment enables the 3D perception of obstacles around autonomous-rail rapid tram, thereby improving the operational reliability in complex urban environments. The test results based on the autonomous-rail rapid transit dataset and Waymo open dataset show that the system can effectively perceive the complex road scenes in autonomous-rail rapid transit, achieving a final 3D average precision (AP) of 0.53, reasoning a single-frame of image about 56 ms, and meeting the real-time requirements of autonomous-rail rapid tram for the obstacle perception algorithm.

The rapid evolution of autonomous driving technology has spurred deeper exploration into its multifaceted domains. This paper focuses on enhancing the intelligence and safety of autonomous driving systems through the application of computer vision. Specifically, the integration of EIoU loss and Varifocal Loss functions within the YOLOv5 algorithm facilitates the training of higher-quality samples at reduced costs. This enhanced YOLOv5 algorithm is adept at detecting traffic entities and incorporates the Matrix NMS algorithm to further refine the loss function adjustments. Moreover, this study employs the proximal policy optimization algorithm to construct a vehicular behavioral decision model for autonomous driving. This model utilizes cubic polynomial equations to depict lane-changing trajectories and introduces a safety distance rule to mitigate collision risks during autonomous maneuvers. Comparative analyses reveal that the modified YOLOv5 algorithm surpasses its predecessors—YOLOv3 and YOLOv4—in terms of detection accuracy and processing speed. Notably, the improved YOLOv5 algorithm exhibits lower network loss values than its original version, achieving faster and more stable convergence. In practical applications, the algorithm successfully identified and labeled 20,025 intelligent vehicular targets, with initial testing accuracy reaching 96.12%. This accuracy improved to 96.36% following the application of EIoU adaptive tuning to reduce risk-free class loss, and further fine-tuning elevated the accuracy to 96.54%. The integration of computer vision into autonomous driving technology thus holds significant promise for advancing the field.

With the increasing application of industrial robots, more and more simple and repetitive operations in industrial automated production lines are replaced by industrial robots. In this study, a path planning algorithm is proposed for the autopilot task of an operating robot to find the optimal path from the start point to the goal point under the constraints of safe obstacle avoidance, shortest distance, and fastest time. Forward kinematics is used to detect whether it will collide with environmental obstacles until the first feasible result is detected as the end position for navigation. Serialized monitoring is used to obtain the activity information of each computational node, the hardware state information of the system, and the state of the autopilot task processing. Finally, obstacle avoidance navigation experiments on a mobile operating robot are conducted to verify the effectiveness and feasibility of the obstacle avoidance navigation algorithm. The results show that the maximum error of the autopilot task of the operating robot is 2.91cm, and the average error is 0.13 2cm, which is in line with the trajectory tracking error requirements of the operating machine control method and verifies the validity and practicability of this study.

The objective of vehicle search is to locate and identify vehicles in uncropped, real-world images, which is the combination of two tasks: vehicle detection and re-identification (Re-ID). As an emerging research topic, vehicle search plays a significant role in the perception of cooperative autonomous vehicles and road driving in the distant future and has become a trend in the future development of intelligent driving. However, there is no suitable dataset for this study. The Tsinghua University DAIR-V2X dataset is utilized to create the first cross-camera vehicle search dataset, DAIR-V2XSearch, which combines the cameras at both ends of the vehicle and the road in real-world scenes. The primary purpose of the current search network is to address the pedestrian issue. Due to varying task scenarios, it is necessary to re-establish the network in order to resolve the problem of vast differences in different perspectives caused by vehicle searches. A phased feature extraction network (PFE-Net) is proposed as a solution to the cross-camera vehicle search problem. Initially, the anchor-free YOLOX framework is selected as the backbone network, which not only improves the network's performance but also eliminates the fuzzy situation in which multiple anchor boxes correspond to a single vehicle ID in the Re-ID branch. Second, for the vehicle Re-ID branch, a camera grouping module is proposed to effectively address issues such as sudden changes in perspective and disparities in shooting under different cameras. Finally, a cross-level feature fusion module is designed to enhance the model's ability to extract subtle vehicle features and the Re-ID's precision. Experiments demonstrate that our proposed PFE-Net achieves the highest precision in the DAIR-V2XSearch dataset. [ABSTRACT FROM AUTHOR]

Higher standards for reliability and efficiency apply to the connection between vehicle terminals and infrastructure by the fifth-generation mobile communication technology (5G). A vehicle-to-infrastructure system uses a communication system called NR-V2I (New Radio-Vehicle to Infrastructure), which uses Link Adaptation (LA) technology to communicate in constantly changing V2I to increase the efficacy and reliability of V2I information transmission. This paper proposes a Double Deep Q-learning (DDQL) LA scheduling algorithm for optimizing the modulation and coding scheme (MCS) of autonomous driving vehicles in V2I communication. The problem with the Doppler shift and complex fast time-varying channels reducing the reliability of information transmission in V2I scenarios is that they make it less likely that the information will be transmitted accurately. Schedules for autonomous vehicles using Space Division Multiplexing (SDM) and MCS are used in V2I communications. To address the issue of Deep Q-learning (DQL) overestimation in the Q-Network learning process, the approach integrates Deep Neural Network (DNN) and Double Q-Network (DDQN). The findings of this study demonstrate that the suggested algorithm can adapt to complex channel environments with varying vehicle speeds in V2I scenarios and by choosing the best scheduling scheme for V2I road information transmission using a combination of MCS. SDM not only increases the accuracy of the transmission of road safety information but also helps to foster cooperation and communication between vehicle terminals to realize cooperative driving. [ABSTRACT FROM AUTHOR]

The automatic driving vehicle has to enter the mixed traffic signal free intersection of people and vehicles. It is necessary to ensure the interaction safety between pedestrians. In order to solve this problem, the non self driving vehicle was taken as the research object, and its interaction with pedestrians at no traffic signal intersection was explored. Two intersections without traffic lights in Beijing were selected as the research scene for long-term shooting. Based on the video data, the individual attribute variables of pedestrians, pedestrian crossing behavior variables, vehicle crossing behavior variables and gap data were extracted, and the pedestrian crossing was divided into two stages before and during crossing. The results show that the crossing stage has a significant impact on the crossing time of pedestrians, and the crossing intersection has no significant impact on the crossing time of pedestrians. For the waiting time before crossing, the factors affecting it in the case of right turn lane are the number of vehicles rejected by pedestrians and the direction from which they come, and the factors affecting it in the case of no right turn lane are the number of vehicles rejected by pedestrians and the starting position. For the crossing time in crossing, the factors affecting it with or without right turn lane are the number of crossing people and the behavior of vehicle 1. In the future, when driving a vehicle to a traffic signal intersection, the result can be used to identify pedestrians and determine the travel time of traveller, so as to make corresponding measures in time. [ABSTRACT FROM AUTHOR]

Over the past ten years, autonomous driving has garnered a great deal of interest from both the scientific community and business. Strong technological advancements have made automated driving more practical because human driving abilities seem limited in terms of driving experience, reaction time, and the effectiveness of real‐time decisions. The development of highly autonomous driving algorithms is inextricably tied to planning and changing a vehicle path that must be user‐acceptable, efficient, and collision‐free. Path planning for road vehicles is a difficult problem due to the high speed involved and the requirement to assure passenger safety. Here, a new path‐planning method is developed for both connected and disconnected automatic road vehicles on multilane highways. This paradigm states that the right phrases to describe the objectives of vehicle improvement, passenger comfort, prevention of vehicle‐to‐vehicle collisions and road deviations are included in the objective function. Hunger Games improved Archimedes optimization (HGE‐ARCO) is used to optimize the paths for achieving better‐planned outcomes. At the 100th penetration rate, the HGE‐ARCO scheme reached a top speed of about 99 km/h. The results shows unmistakably that the proposed HGE‐ARCO produces a time of 12.3021 s, which is less than other conventional methods. [ABSTRACT FROM AUTHOR]

A path tracking system is required to fully meet the change of operating speed in the agricultural machinery. In this study, a predictive control system was proposed using a preview kinematics model. The model of preview heading error was also established using the preview following theory. The output equation was combined with the conventional state equation of path tracking error. The state space model of path tracking error was then constructed for the preview kinematic after linearization and discretization. Then the path tracking control was designed using a model predictive control algorithm and input parameterized attenuation. The co-simulation platform was built for the path-tracking control of agricultural machinery using CarSim and Simulink software. The simulation results showed that the lateral errors of straight path tracking asymptotically approached zero for the preview model predictive controller. There was no overshoot of driving curves at different operating speeds. Once the operating speeds of agricultural machinery were 1, 3 and 5 m/s, respectively, the maximum absolute values of lateral error were 8.52, 10.42, and 10.82 cm in the circular path tracking for the preview model predictive controller, whereas, the standard values of lateral error were 3.96, 5.83 and 6.17 cm, respectively. Once the control horizon layers were 10, 30 and 60, the operation period of the preview model predictive controller was reduced by 7.5%, 43.0% and 48.5%, respectively, compared with the conventional. The preview model predictive controller was used to dynamically adjust the preview distance and prediction horizon, according to the operating speed and path curvature, respectively. The dynamic performance of the path tracking system was effectively improved with the control accuracy. There was more balanced path tracking at different operating speeds, indicating the simpler parameter setting. The decay factor was also introduced to parameterize the control input. The amount of online optimization was greatly reduced to improve the real-time performance of path-tracking control. The field test platform of automatic driving was built to install the satellite antenna, satellite receiver, angle sensor, proportional reversing valve, controller, and vehicle touch screen on the agricultural machinery. In straight path tracking, the average maximum absolute values of lateral error were 3.50 and 5.32 cm, the average absolute values of lateral error were 1.46 and 1.93 cm, and the average standard values of lateral error were 1.71 and 2.38 cm, respectively, when the operating speeds of agricultural machinery were 1 and 5 m/s, respectively. In circular path tracking, the average maximum absolute values of lateral error were 9.62 and 15.34 cm, the average absolute values of lateral error were 5.17 and 7.86 cm, and the average standard values of lateral error were 4.82 and 8.39 cm, respectively, when the operating speeds of agricultural machinery were 1 and 5 m/s, respectively. The field test results showed that the preview model predictive controller shared a strong adaptability to the change of operating speed within the operating speed range of 0.5 to 5 m/s. Agricultural machinery quickly and stably tracked the reference path with high control accuracy. The average maximum absolute values and the average standard values of lateral error were less than 5.5 and 2.5 cm, respectively, for the straight path tracking, while less than 15.5 and 8.5 cm, respectively, for the circular path tracking. The preview model predictive controller can fully meet the actual control requirements of agricultural machinery, especially suitable for complex environments or high-speed conditions. [ABSTRACT FROM AUTHOR]

In recent years, with the rapid development of artificial intelligence technology, more and more companies have shifted their attention to the field of automatic driving. The target detection is one of the core technologies in the field of automatic driving, which can perceive the environment by using distributed sensors and affect the vehicle driving control. How to describe vehicle image is particularly important in target detection for automatic driving. In this paper, we combine edge information, HOG features and local association features to describe the targets in image for automatic driving. Then, the extracted features are used to learn a classifier for further vehicle detection. The experimental results show that the proposed method can achieve better prevision, recall, and F1‐measure than merely using edge features or HOG features. The proposed method can extract more comprehensive features for vehicle detection. [ABSTRACT FROM AUTHOR]

Automatic driving technology refers to equipment such as vehicle-mounted sensors and computers that are used to navigate and control vehicles autonomously by acquiring external environmental information. To achieve automatic driving, vehicles must be able to perceive the surrounding environment and recognize and understand traffic signs, traffic signals, pedestrians, and other traffic participants, as well as accurately plan and control their path. Recognition of traffic signs and signals is an essential part of automatic driving technology, and gesture recognition is a crucial aspect of traffic-signal recognition. This article introduces mm-TPG, a traffic-police gesture recognition system based on a millimeter-wave point cloud. The system uses a 60 GHz frequency-modulated continuous-wave (FMCW) millimeter-wave radar as a sensor to achieve high-precision recognition of traffic-police gestures. Initially, a double-threshold filtering algorithm is used to denoise the millimeter-wave raw data, followed by multi-frame synthesis processing of the generated point cloud data and feature extraction using a ResNet18 network. Finally, gated recurrent units are used for classification to enable the recognition of different traffic-police gestures. Experimental results demonstrate that the mm-TPG system has high accuracy and robustness and can effectively recognize traffic-police gestures in complex environments such as varying lighting and weather conditions, providing strong support for traffic safety. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Rapid developments in automatic driving technology have given rise to new experiences for passengers. Safety is a main priority in automatic driving. A strong familiarity with road-surface conditions during the day and night is essential to ensuring driving safety. Existing models used for recognizing road-surface conditions lack the required robustness and generalization abilities. Most studies only validated the performance of these models on daylight images. To address this problem, we propose a novel multi-supervised bidirectional fusion network (MBFN) model to detect weather-induced road-surface conditions on the path of automatic vehicles at both daytime and nighttime. We employed ConvNeXt to extract the basic features, which were further processed using a new bidirectional fusion module to create a fused feature. Then, the basic and fused features were concatenated to generate a refined feature with greater discriminative and generalization abilities. Finally, we designed a multi-supervised loss function to train the MBFN model based on the extracted features. Experiments were conducted using two public datasets. The results clearly demonstrated that the MBFN model could classify diverse roadsurface conditions, such as dry, wet, and snowy conditions, with a satisfactory accuracy and outperform state-of-the-art baseline models. Notably, the proposed model has multiple variants that could also achieve competitive performances under different road conditions. The code for the MBFN model is shared at https://zenodo. org/badge/latestdoi/607014079. [ABSTRACT FROM AUTHOR]

Copyright of China Mechanical Engineering is the property of Editorial Board of China Mechanical Engineering and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

As intelligent driving vehicles came out of concept into people’s life, the combination of safe driving and artificial intelligence becomes the new direction of future transportation development. Autonomous driving technology is developing based on control algorithms and model recognitions. In this paper, a cloud-based interconnected multi-sensor fusion autonomous vehicle system is proposed that uses deep learning (YOLOv4) and improved ORB algorithms to identify pedestrians, vehicles, and various traffic signs. A cloud-based interactive system is built to enable vehicle owners to master the situation of their vehicles at any time. In order to meet multiple application of automatic driving vehicles, the environment perception technology of multi-sensor fusion processing has broadened the uses of automatic driving vehicles by being equipped with automatic speech recognition (ASR), vehicle following mode and road patrol mode. These functions enable automatic driving to be used in applications such as agricultural irrigation, road firefighting and contactless delivery under new coronavirus outbreaks. Finally, using the embedded system equipment, an intelligent car was built for experimental verification, and the overall recognition accuracy of the system was over 96%.

Aiming at the problems of punctuality, parking accuracy and energy saving of urban rail train operation, an intelligent control method for automatic train operation (ATO) based on deep Q network (DQN) is proposed. The train dynamics model is established under the condition of satisfying the safety principle and various constraints of automatic driving of urban rail train. Considering the transformation rules and sequences of working conditions between train stations, the agent in the DQN algorithm is used as the train controller to adjust the train automatic driving strategy in real time according to the train operating state and operating environment, and optimizes the generation of the train automatic driving curve. Taking the Beijing Yizhuang Subway line as an example, the simulation test results show that the DQN urban rail train control method reduces energy consumption by 12.32% compared with the traditional train PID control method, and improves the running punctuality and parking accuracy; at the same time, the DQN train automatically driving control method can adjust the train running state in real time and dynamically, and has good adaptability and robustness to the change of train running environment parameters.

As one of the important way for constructing smart agriculture, unmanned farms are the most attractive in nowadays, and have been explored in many countries. Generally, data, knowledge and intelligent equipment are the core elements of unmanned farms. It deeply integrates modern information technologies such as the Internet of Things, big data, cloud computing, edge computing, and artificial intelligence with agriculture to realize agricultural production information perception, quantitative decision-making, intelligent control, precise input and personalized services. In the paper, the overall technical architecture of unmanned farms is introduced, and five kinds of key technologies of unmanned farms are proposed, which include information perception and intelligent decision-making technology, precision control technology and key equipment for agriculture, automatic driving technology in agriculture, unmanned operation agricultural equipment, management and remote controlling system for unmanned farms. Furthermore, the latest research progress of the above technologies both worldwide are analyzed. Based on which, critical scientific and technological issues to be solved for developing unmanned farms in China are proposed, include unstructured environment perception of farmland, automatic drive for agriculture machinery in complex and changeable farmland environment, autonomous task assignment and path planning of unmanned agricultural machinery, autonomous cooperative operation control of unmanned agricultural machinery group. Those technologies are challenging and absolutely, and would be the most competitive commanding height in the future. The maize unmanned farm constructed in the city of Gongzhuling, Jilin province, China, was also introduced in detail. The unmanned farms is mainly composed of information perception system, unmanned agricultural equipment, management and controlling system. The perception system obtains and provides the farmland information, maize growth, pest and disease information of the farm. The unmanned agricultural machineries could complete the whole process of the maize mechanization under unattended conditions. The management and controlling system includes the basic GIS, remote controlling subsystem, precision operation management subsystem and working display system for unmanned agricultural machineries. The application of the maize unmanned farm has improved maize production efficiency (the harvesting efficiency has been increased by 3－4 times) and reduced labors. Finally, the paper summarizes the important role of the unmanned farm technology were summarized in solving the problems such as reduction of labors, analyzes the opportunities and challenges of developing unmanned farms in China, and put forward the strategic goals and ideas of developing unmanned farm in China.