Highly accurate perception is one of the pivotal factors for the safe operation of Intelligent Connected Vehicles (ICVs). Nevertheless, occlusion blind spots, limited fields-of-view, and low-point density of the senso...
Highly accurate perception is one of the pivotal factors for the safe operation of Intelligent Connected Vehicles (ICVs). Nevertheless, occlusion blind spots, limited fields-of-view, and low-point density of the sensor data lead to limited perception for the single ICV, which can be well addressed with vehicle-to-vehicle (V2V) cooperative perception. Recent development of V2V cooperative perception technology have made the perception of ICVs more and more accurate and reliable. In V2V cooperative perception, LiDAR and Camera are two types of complementary sensors for ICVs. However, using only specific single-modal data such as Camera RGB images or LiDAR point clouds for V2V collaborative perception cannot fully improve perception accuracy. Furthermore, the large model based on Transformers has been proven to effectively enhance multi-modal fusion. To this end, we propose MCoT, a novel approach for multi-modal V2V cooperative perception with Transformers. Our MCoT extracts intermediate features from RGB images and point cloud of different agents, aligning them in the Bird’s-Eye View (BEV) perspective through rigid association. Subsequently, we use the cross-attention mechanism to perform a soft fusion of these features in the BEV domain. The attention mechanism empowers our model with the ability to adaptively discern which regions of the image and LiDAR are most relevant, and what information should be extracted from them. Extensive evaluations demonstrate that MCoT can significantly enhance the accuracy and robustness of perception. Our model achieved remarkable results on the large-scale simulation dataset OPV2V, improving the average accuracy by 71.43% compared to the baseline and outperforming the second-place by nearly 3.95%. Our approach also demonstrates the fastest convergence rate under the same number of training epochs.
Cloud systems are managed on the basis of au-tonomous systems. We present criteria that are suitable for optimization or improvement of modern cloud systems. In every operating system, task scheduling is very importan...
详细信息
Cloud systems are managed on the basis of au-tonomous systems. We present criteria that are suitable for optimization or improvement of modern cloud systems. In every operating system, task scheduling is very important. In clouds systems, where large amount of tasks runs on numerous machines, optimized task scheduling leads to significant reduction of computing time. Cloud providers have to comply with service level agreement from technical and from the quality of service point of view. For this reason, we specify quality of service criteria and limits for service level agreement violation. Clouds are virtualized by virtual machines or containers. We show approaches for power consumption minimization. Fog computing helps to improve middleware technology between cloud computing and the IoT devices. We specify criteria for correct decomposition of parallel and distributed application. We conclude that understanding of effective work of the system, can improve design and implementation of parallel and distributed application.
The proceedings contain 49 papers. The special focus in this conference is on parallel and distributedcomputing, Applications and Technologies. The topics include: A real-time routing protocol in wireless sensor-actu...
ISBN:
(纸本)9789811359064
The proceedings contain 49 papers. The special focus in this conference is on parallel and distributedcomputing, Applications and Technologies. The topics include: A real-time routing protocol in wireless sensor-actuator network;privacy preserving classification based on perturbation for network traffic;fault diagnosis of a wireless sensor network using a hybrid method;an optimization theory of home occupants’ access data for determining smart grid service;automatic classification of transformed protocols using deep learning;covert timing channel design for uniprocessor real-time systems;parallelization of the DIANA algorithm in openMP;Flash animation watermarking algorithm based on SWF tag attributes;efficient scheduling strategy for data collection in delay-tolerant wireless sensor networks with a mobile sink;analysis of massive e-learning processes: An approach based on big association rules mining;SGNet: Design of optimized DCNN for real-time face detection;A study on L1 data cache bypassing methods for high-performance GPUs;Memory contention aware power management for high performance GPUs;Dynamic selective warp scheduling for GPUs using L1 data Cache locality information;an efficient model and algorithm for privacy-preserving trajectory data publishing;what makes charitable crowdfunding projects successful: A research based on data mining and social capital theory;A SwarmESB based architecture for an european healthcare insurance system in compliance with GDPR;a study on deriving and simulating pre-risk on complex gas facilities for preventing accidents;body gesture modeling for psychology analysis in job interview based on deep spatio-temporal approach;green vs revenue: Data center profit maximization under green degree constraints;evaluation for two bloom filters’ configuration.
To alleviate the losses caused owing to the sporadic nature of renewable power sources to the distributed generation system and to make use of these resources in a more efficient way, a fuzzy logic control based pursu...
详细信息
As power systems continue to evolve, aggregation technology has become a mainstream solution. This article aims to provide a detailed analysis and study of data acquisition issues in centralized transmission lines. In...
As power systems continue to evolve, aggregation technology has become a mainstream solution. This article aims to provide a detailed analysis and study of data acquisition issues in centralized transmission lines. In the aggregation of transmission lines, this article adopts a distributedcomputing service model, which is to set up a centralized municipal area network through the network to realize the monitoring and management of load demand and power supply reliability at the bureau and county level. This article combines smart terminals with traditional networks to meet users' different application functional requirements, and uses the characteristics of the Internet to provide a data collection platform with strong realtime performance, high reliability, and large maintenance and management capabilities. Finally, this article tests the performance of the platform. The test results show that the system receives data quickly and processes data quickly. The overall disk utilization index is above 0.84, and the highest index of 0.98 is infinitely close to 1, which shows that the performance of the system is efficient.
Maintaining the integrity of programs is crucial for ensuring the security of the platform they operate on. The integrity of the programs on a trusted computing platform is measured and reported to a challenger by uti...
Maintaining the integrity of programs is crucial for ensuring the security of the platform they operate on. The integrity of the programs on a trusted computing platform is measured and reported to a challenger by utilizing a Trusted Platform Module integrated on the platform. The challenger then judges whether the programs maintain their integrity by comparing the measurements with the integrity references. Therefore, managing the integrity references and ensuring their correctness is of utmost importance. There is currently no system where a challenger can obtain all integrity references directly. A challenger typically has to obtain the integrity references from each platform with “good” measurements. This paper presents BIRM, a consortium Blockchain-based Integrity References Management system. BIRM enforces access controls for publishing integrity references. The integrity reference files in BIRM use a unified syntax based on the JSON Web Signature specification and are stored separately in a distributed storage network. Only identity and storage index tuples are recorded in the blockchain to reduce its load. The blocks in BIRM are organized as a Merkle Mountain Range. When an inquiry is made, BIRM not only returns the requested identity and storage index tuple directly from the key-value database but also generates proof data by parsing block data to ensure the tuple's correctness. BIRM supports multiple tuple inquiries in a single request, using multiple paths in the Merkle Mountain Range tree. The amount of proof data and proof generation and checks are logarithmic in complexity relative to the total number of blocks.
This paper investigates the lightning performance on 10kV overhead distribution lines grounded with counterpoise wires under extremely low soil conductivity. A partial element equivalent circuit model considering loss...
详细信息
ISBN:
(纸本)9781665423465
This paper investigates the lightning performance on 10kV overhead distribution lines grounded with counterpoise wires under extremely low soil conductivity. A partial element equivalent circuit model considering lossy ground is used for surge analysis, and the Monte Carlo method is adopted for evaluating the flashover rate. It is found that under extremely low soil conductivity, the simple vertical grounding rod is not effective in lightning protection, while the performance is significantly increased with extended counterpoise wires. The single-row counterpoise wire with a length of pole span performs best among the schemes of concern. The single-row counterpoise wire with a half-length of pole span is a competitive one, due to a half-cost reduction but a competitive performance with a difference of less than 9%. With the same total wire length, the scheme of single-row counterpoise wire is slightly better than the parallel-row arrangement. If a higher lightning protection level is required, the improvement could be achieved by increasing the CFO level or shortening the span of two adjacent surge arresters, since increasing the length of counterpoise wire is not cost-effective.
Electric power data contains rich private information, and encryption technology is often used to protect its information security. Due to its huge data volume, considering the lightweight and high-efficiency encrypti...
详细信息
Electric power data contains rich private information, and encryption technology is often used to protect its information security. Due to its huge data volume, considering the lightweight and high-efficiency encryption algorithm, based on the encryption algorithm SM4 with independent intellectual property rights in my country, combined with the big data technology Hadoop, a parallel encryption speed-up scheme based on the national encryption algorithm SM4 is designed to realize the data encryption speed. Encryption and decryption protection and acceleration of computation. Finally, an experimental platform based on Hadoop is built, and the data is encrypted in blocks with the MapReduce computing framework, and the feasibility test of the scheme is carried out. The experimental results show that the algorithm scheme can improve the encryption speed of unstructured data files. The optimal number of Maps in the established cluster is 8, and the speedup ratio between serial encryption and parallel encryption is more than 2. Compared with serial encryption, it can save 50% of computing overhead time.
The proceedings contain 11 papers. The special focus in this conference is on Languages and Compilers for parallelcomputing. The topics include: CubeGen: Code Generation for Accelerated GEMM-Based Convolution with Ti...
ISBN:
(纸本)9783030727888
The proceedings contain 11 papers. The special focus in this conference is on Languages and Compilers for parallelcomputing. The topics include: CubeGen: Code Generation for Accelerated GEMM-Based Convolution with Tiling;PostSLP: Cross-Region Vectorization of Fully or Partially Vectorized Code;FLARE: Flexibly Sharing Commodity GPUs to Enforce QoS and Improve Utilization;foundations of Consistency Types for a Higher-Order distributed Language;Common Subexpression Convergence: A New Code Optimization for SIMT Processors;Using Performance Event Profiles to Deduce an Execution Model of MATLAB with Just-In-Time Compilation;CLAM: Compiler Leasing of Accelerator Memory;abstractions for Polyhedral Topology-Aware Tasking [Position Paper];SWIRL ++ : Evaluating Performance Models to Guide Code Transformation in Convolutional Neural Networks.
The proceedings contain 88 papers. The topics discussed include: a high-precision lens distortion correction method for vision measurement;intelligent data security threat discovery model based on grid data;building e...
ISBN:
(纸本)9781665443685
The proceedings contain 88 papers. The topics discussed include: a high-precision lens distortion correction method for vision measurement;intelligent data security threat discovery model based on grid data;building extraction based on U-net and conditional random fields;the application of artificial intelligence on the traceability and dispersion of air pollution;video captioning in compressed video;constrained policy optimization algorithm for autonomous driving via reinforcement learning;research on classification of wild fungi based on improved Resnet50 network;parallel approach to high-precision smoke simulation based on a multi-GPU framework;design of a global retrieval system for characteristic data based on SOLR;and an image enhancement method in coal mine underground based on deep retinex network and fusion strategy.
暂无评论