High accuracy vehicle localization information is critical for intelligent transportation systems and future autonomous vehicles. It challenging to achieve the required centimeter-level accuracy, especially in urban or global navigation satellite system denied environments. Here we propose a vehicle-to-infrastructure (V2I)-based algorithm. First, it low-cost hardware requirements are simplified, minimum requirement single roadside unit on-board receiver. Second, computationally efficient,...

Due to the small size, high density, and background noise associated with strip surface defects, current object detection model commonly faces limitations in performance. To address this issue, we propose a spatial-to-depth feature-enhanced method called STD-Detector. The consists of two types STD-Conv-A STD-Conv-B. First, module is used backbone feature extraction network expand field perception enable learn wider range information. Then, STD-Conv-B for fusion networks improve expression...

Accurate vehicle localization is critical for connected and autonomous vehicles of the future. But it challenging, especially in global positioning system (GPS) denied or urban environments. In this paper, we propose an efficient GPS-free algorithm by exploiting vehicle-to-infrastructure communications. It on-board odometer inertial measurement unit (IMU) assisted single roadside (RSU) based approach. The available RSU position, speed vector distance information between are formulated as...

Community detection is a critical task for complex network analysis. It helps us to understand the properties of system that represents and has significance wide range applications. Nowadays, challenges faced by community algorithms include overlapping structure detection, large scale analysis, dynamic changing analyzed topology many more. In this paper self-organizing algorithm, based on idea swarm intelligence, was proposed its parallel algorithm designed Giraph++ which semi-asynchronous...

Deeplab series semantic segmentation algorithms extract target features using deep layers of a convolutional neural network, resulting in lacking detailed information, such as edges and shapes extracted by shallow layers. Deeplabv3plus uses atrous convolution to obtain feature maps, which lose some image information. All the above have an impact on performance improvement. In response these issues, reduce performance, we propose algorithm based multi-expert system that builds multiple expert...

Aiming at the problem that traditional IoU-NMS algorithm has poor filtering of redundant boxes with high confidence scores in RetinaNet, an object detection based on cosine similarity IoU is proposed. Based original calculation method, between calculated by vector to better evaluate them, which removes and retain more accurate boxes. Meanwhile, CBAM added ResNet-50 network extract richer semantic information. accuracy PASCAL VOC dataset reaches 80.6%, 2.1% higher than benchmark algorithm.

In order to overcome the problem of over-segmentation caused by traditional watershed algorithm, a color image segmentation algorithm based on simple linear iterative clustering (SLIC) and is proposed achieve an ideal effect. Firstly, calculates number super-pixels pre-segmented complexity, uses SLIC super-pixel preprocessing original reduce redundant information in subsequent processing. Then, adaptive threshold calculation method process gradient preprocessed effectively remove noise...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

This paper investigates the impact of securities relending on China stock prices. As was approved in February 28, 2013, it provides a very unique data set to test relending. With event study methodology, we find that prices exhibits negative since cumulative abnormal return after date is negative. Cross-sectional regression shows negatively correlated turnover and volatility, positively not significantly trading volume, price earning ratio, book ratio.

Recently, although Transformer has widespread application in the field of computer vision, quadratic complexity its Self-Attention hindered processing large-scale image captioning tasks. Therefore, this paper, we propose a Learnable Linear-Attention with Fast-Normalization for Large-Scale Image Captioning (dubbed as LLAFN-Generator). Firstly, it introduces (LLA) module to solve weight score learning images, which is simply implemented through two linear layers and greatly reduces computation...

目的 针对现有视频目标分割（video object segmentation，VOS）算法不能自适应进行样本权重更新，以及使用过多的冗余特征信息导致不必要的空间与时间消耗等问题，提出一种自适应权重更新的轻量级视频目标分割算法。方法 首先，为建立一个具有较强目标判别性的算法模型，所提算法根据提取特征的表征质量，自适应地赋予特征相应的权重；其次，为了去除冗余信息，提高算法的运行速度，通过优化信息存储策略，构建了一个轻量级的记忆模块。结果 实验结果表明，在公开数据集DAVIS2016 （densely annotated video segmentation）和DAVIS2017上，本文算法的区域相似度与轮廓准确度的均值J&F分别达到了85.8%和78.3%，与对比的视频目标分割算法相比具有明显的优势。结论 通过合理且无冗余的历史帧信息利用方式，提升了算法对于目标建模的泛化能力，使目标掩码质量更高。;Objective Video segmentation is a basic computer vision task that widely used in editing,...

为了解决相关滤波视觉跟踪算法在复杂场景中产生的跟踪漂移问题, 提出一种融合检测机制的相关滤波跟踪框架。利用时空正则化滤波器作为跟踪器, 同时使用线性核相关滤波器作为检测器。当跟踪器与目标进行相关计算得到的响应图为多个峰值时, 激活检测器, 对多个峰值进行相关匹配, 获得重检测结果; 同时, 使用平均峰值相关能量的滤波器模型更新策略得到更加可靠的检测器, 以达到提高跟踪精度和算法鲁棒性的目的。在OTB2015、Temple color 128和VOT2016数据平台上的实验结果表明, 与近年提出的性能较出色的跟踪算法相比, 本文算法在目标运动模糊、相似背景干扰和光照变化等复杂场景中具有更好的鲁棒性和准确性, 且跟踪精度和成功率上均有提高。

At present, several mainstream algorithms using color name (CN) all adopt Principal Component Analysis (PCA) to process the feature. However, PCA assumes that noise of input data must obey Gaussian distribution, which is a conspicuous defect. Aim address this problem, in paper, we take Robust (Robust PCA) CN features. The method projects original RGB space robust space–CN space, means image stratified 11 layers according name. Then, it processes features by PCA, so mapped information...

With increasing demand of high purity vanadium pentoxide (V2O5) for manufacturing all-vanadium redox flow batteries (VRFBs) in large-scale power storage, production V2O5 by chlorination process is an efficient and clean approach, which oxidation oxytrichloride (VOCl3) oxygen the controlling step. In present work, gaseous VOCl3 deposition behavior on particle substrates a fluidized bed reactor was investigated. Results showed that thermodynamic equilibrium conversion gradually decreased with...

To fully develop the complementary advantages of different visual features and to improve robustness multi-feature fusions, we propose a robust correlation filter tracker with adaptive multi-complementary fusion based on game theory. By combining selected from handcrafted convolution features, our method constructs two combined in tracking framework discriminative filters (DCFs). In addition, by utilizing theory, are regarded as sides game, achieving best balance through continuous gaming...