In this paper, we explore the use of cooperative perception to improve the perception performance of autonomous vehicles. By aggregating perception data from multiple nearby vehicles and roadside units, we can see through obstructions, detect objects at a distance, and improve detection accuracy. We propose a new cooperative perception framework, V2X-HAN. This framework mainly uses a heterogeneous graph attention network model, which can better capture the complex structure and rich information in the heterogeneous graph, achieve better feature fusion, and thus improve the accuracy of detection. We trained and validated the OPV2V and V2XSet datasets, and compared them with related models. Many experimental results show that V2X-HAN has achieved good detection results in cooperative perception.
Driving safety has always been a core concern in the field of Autonomous driving vehicles, and the safety issue of interaction between the ego-vehicle and the surrounding vehicles is particularly important in lane-changing scenarios. This paper proposes a deep deterministic policy gradient algorithm (DDPG) based on driving risk for obstacle avoidance decision-making model (RISK-DDPG), which aims to learn risk-aware driving decision-making strategies at a minimized cost. The DDPG algorithm is used to learn the minimally risky driving strategy by using Gaussian risk field to assess the magnitude of the risk of obstacles and the value of the driving risk quantum as a reward function. The performance of the proposed algorithm in this paper is evaluated against traditional obstacle avoidance algorithms by evaluating the performance in the CARLA simulation environment. The results show that the proposed model achieves safer obstacle avoidance decisions while speeding up the model convergence.
Cracks are one of the most common damages in highway bridges, and timely detection of existing cracks is crucial for the safety of highway bridges. Targeting the issues of low crack detection accuracy and many missed detections in complex backgrounds, a new network ES-YOLOX (Embed-Shuffle-YOLOX) is proposed based on the YOLOX object detection algorithm. This method enhances the backbone network's capability for feature extraction by embedding additional CSPLayer, introducing the reconfigured ShuffleNetV2 structure, introducing depth-separable convolution to decrease both the parameter count and computational load of the network, and finally introducing the ECA attention module to make the network more concerned with crack features. Experimental results show that the proposed method increases the mAP50 value by 2.11 % and the F1 score by 2.17 %, and reduces the number of parameters and calculations by 55.8% and 62.7%.
A vehicle's lane-changing behavior is affected by the surrounding environment and driver factors, which makes it difficult to identify accurately. To solve this problem, a personalized lane-changing decision model based on a Long Short-term Memory (LSTM) network is proposed. First, an unsupervised clustering method is applied to recognize three distinct driving styles; Second, by considering the interactions among the target vehicle and surrounding vehicles, a benefit function is constructed to measure these interactions and generate the lane-changing benefit values. The lane-changing gain values and feature parameters are used as model inputs to construct a personalized lane-changing decision model using LSTM. Finally, the proposed method is validated with the NGSIM dataset: the overall accuracy of the model reaches 97.8% when considering different driving styles, which proves that the proposed method can achieve personalized lane-changing decisions based on different drivers' lane-changing behaviors.
Mobile edge computing (MEC) is a promising paradigm for offloading compute-intensive services on vehicles to alleviate the problem of limited resources in the vehicles themselves. However, since the vehicle network involves multiple edge servers, MEC is facing the dilemma of how to fully utilize the edge resources to achieve the maximum benefit. In this paper, we aim to analyze MEC task offloading strategies from a multi-objective optimization perspective by considering independent partitionable computational tasks, and an MEC communication and computation offloading framework is constructed. A partial computational resource optimization (PCRO) algorithm is proposed, which jointly considers computational resource allocation and unit price adjustment to achieve minimum cost for vehicle users and maximum profit for edge servers. Extensive experimental results demonstrate the effectiveness of our proposed PCRO algorithm.
The lane line detection and recognition are crucial research area for automatic driving. It aims at solving the problem of fuzzy feature expression and low time-sensitives of lane line detection based on semantic segmentation. This paper proposes to remove irrelevant background by dynamic programming region of interest while improving the lightweight neural network (U-Net). A group-by-group convolution and depth wise separable convolution in the backbone network are introduced, simplifies the branches of the backbone network, and atrous convolution is introduced into the enhanced path network with multi-level skip connection structure to retain the underlying coarse-grained semantic feature information. The full-scale skip connection fusion mechanism of the decoder is preserved, while capturing the fine-grained semantics and coarse-grained semantics of the feature map at full scale. The introduction of skip connections between the decoder and the encoder can enhance the lanes without increasing the size of the receptive field. The ability to extract line features and the ability to extract context improves the accuracy of lane lines. The experimental results show that the improved neural network can obtain good detection performance in complex lane lines, and effectively improve the accuracy and time-sensitives of lane lines.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.