三维模型边缘特征与异常检测

针对三维模型识别和检测问题,提出一种新的基于边缘特征的三维模型异常检测方法。将每一个三维模型利用边缘特征表示为一条时间序列,对产生的时间序列集进行Isodata聚类,利用聚类结果经过两次划分实现异常检测。第一次划分过程产生候选异常和候选正常,第二次划分过程在候选异常中进一步选出检测结果。实验结果表明,该算法性能优于传统的基于距离、邻近度以及基于相对密度的异常检测算法,在一定条件下,也优于基于密度的异常检测算法。     

Computer-aided detection and classification of microcalcifications in mammograms: a survey

Breast cancer continues to be a significant public health problem in the world. Approximately, 182,000 new cases of breast cancer are diagnosed and 46,000 women die of breast cancer each year in the United States. Even more disturbing is the fact that one out of eight women in US will develop breast cancer at some point during her lifetime. Primary prevention seems impossible since the causes of this disease still remain unknown. Early detection is the key to improving breast cancer prognosis. Mammography is one of the reliable methods for early detection of breast carcinomas. There are some limitations of human observers, and it is difficult for radiologists to provide both accurate and uniform evaluation for the enormous number of mammograms generated in widespread screening. The presence of microcalcification clusters (MCCs) is an important sign for the detection of early breast carcinoma. An early sign of 30–50% of breast cancer detected mammographically is the appearance of clusters of fine, granular microcalcification, and 60–80% of breast carcinomas reveal MCCs upon histological examinations. The high correlation between the appearance of the microcalcification clusters and the diseases show that the CAD (computer aided diagnosis) systems for automated detection/classification of MCCs will be very useful and helpful for breast cancer control. In this survey paper, we summarize and compare the methods used in various stages of the computer-aided detection systems (CAD). In particular, the enhancement and segmentation algorithms, mammographic features, classifiers and their performances are studied and compared. Remaining challenges and future research directions are also discussed.     

利用正交型锁相放大器实现三维磁场微弱信号检测

变磁场测量(ACFM)是近年兴起的可以定量的无损检测技术,为实现该检测方法,设计了由矩形激励线圈和正方体检测线圈构成的三维ACFM传感器激励和感应空间磁场的变化,提出了利用正交型锁相放大器(LIA)的信号检测方法实现传感器的每个方向毫伏级变化的微弱感应电压信号的精确测量。并在此基础上建立了基于频率扫描技术的缺陷识别模型,结果表明:该检测方法达到了对三维磁场微弱感应信号测量和对任意走向裂纹识别的目的。     

伪3D卷积神经网络与注意力机制结合的疲劳驾驶检测

目的 复杂环境下的疲劳驾驶检测是一个具有挑战性的技术问题。为了充分利用驾驶员面部特征信息与时间特征,提出一种基于伪3D（Pseudo-3D,P3D）卷积神经网络（convolutional neural network,CNN）与注意力机制的驾驶疲劳检测方法。方法 采用伪3D卷积模块进行时空特征学习;提出P3D-Attention模块,利用P3D的结构融合双通道注意力模块和适应的空间注意力模块,提高对重要通道特征的相关度,增加特征图的全局相关性,将多层深度卷积特征进行融合。利用双通道注意力模块分别在视频帧之间和每一帧的通道上施加关注,去除背景和噪声对识别的干扰,使用自适应空间注意模块使模型训练更快、收敛更好;使用2D全局平均池化层替代3D全局平均池化层获得更具表达能力的特征,进而提高网络收敛速度;运用softmax分类层进行分类。结果 在公共数据集YawDD（a yawning detection dataset）上开展对比实验,本文方法在测试集上的F1-score检测准确率达到99.89%,在打哈欠类别上召回率达到100%;在数据集UTA-RLDD（University of Texas at Arlington real-life drowsiness dataset）上,本文方法在测试集上的F1-score检测准确率达到99.64%,在困倦类别上召回率达到100%;与Inception-V3融合LSTM（long short-term memory）的方法相比,本文方法模型大小为42.5 MB,是其模型大小的1/9,本文方法预测时间约660 ms,是其11%左右。结论 提出一种基于伪3D卷积神经网络与注意力机制的驾驶疲劳检测方法,利用注意力机制进一步分析哈欠、眨眼和头部特征运动,将哈欠行为与说话行为动作很好地区分开来。     

A method for the analysis of the interaction between users and objects in 3D navigational space

Along with the improvement of eye-tracking technology, more and more distinct field of researches have introduced movements of the eye in relation to the head to understand user behavior. Most of current researches focus on the perception process of single 2-dimensional images by fixed eye-tracking devices or the head-mount devices. A method of applying eye-tracking on the analysis of the interaction between users and objects in 3D navigational space is proposed in this article. It aims to understand the visual stimulation of 3D objects and the user’s spatial navigational reactions while receiving the stimulation, and proposes the concept of 3D object attention heat map. It also proposes to construct a computational visual attention model for different geometric featured 3D objects by applying the method of feature curves. The VR results of this study also provide future assistance in the incoming immersive world. This study sets to promote eye-tracking from the mainstream of 2D field to 3D spaces and points to a deeper understanding between human and artificial product or natural objects. It would also serve an important role in the field of human-computer interaction, product usability, aids devices for cognition degenerative individuals, and even the field of visual recognition of daily human behavior.     

基于定位不确定性的鲁棒3D目标检测方法

针对3D点云训练数据因人工标注不精确而导致模型定位不准确的问题,提出了一种基于定位不确定性的鲁棒3D目标检测方法。首先,以基于3D体素网格的稀疏嵌入卷积检测（SECOND）网络作为基础网络,在候选区域生成网络（RPN）的基础上增加对定位不确定性的预测;然后,在训练过程中使用高斯和拉普拉斯两种分布模型对定位不确定性进行建模,并对定位损失函数进行重新定义;最后,在预测过程中结合定位不确定性和分类置信度作为目标置信度,使用阈值过滤和非极大值抑制（NMS）方法筛选候选目标。实验结果表明,在KITTI 3D目标检测数据集上,所提算法对于车辆类别的检测准确率在中等难度上比SECOND网络提高了0.5个百分点;当在训练数据中额外加入扰动模拟噪声的情况下,所提算法的检测准确率比SECOND网络最多提高了3.1个百分点。所提算法提高了3D目标检测准确率,减少了误检且提高了3D边界框的精度,并且对于带噪声的数据更鲁棒。     

利用动态规划和电势理论规划UAV三维航迹

用改进的电势理论规划无人机的三维航迹。充分考虑了地形威胁、火力及雷达威胁同时存在的情况,并对它们的权重对航迹规划的影响进行了研究。通过限制航线到包含任务起止点直线的距离,不但克服了电势法无法收敛目标点的问题,而且可以缩短飞行航程;研究惩罚函数中各因素不同权重对飞行航线的影响,选择出最优权重系数;通过仿真试验可知,该方法能够成功地为无人机规划出安全的飞行路径。     

结合混合域注意力与空洞卷积的3维目标检测

目的 通过深度学习卷积神经网络进行3维目标检测的方法已取得巨大进展,但卷积神经网络提取的特征既缺乏不同区域特征的依赖关系,也缺乏不同通道特征的依赖关系,同时难以保证在无损空间分辨率的情况下扩大感受野。针对以上不足,提出了一种结合混合域注意力与空洞卷积的3维目标检测方法。方法 在输入层融入空间域注意力机制,变换输入信息的空间位置,保留需重点关注的区域特征;在网络中融入通道域注意力机制,提取特征的通道权重,获取关键通道特征;通过融合空间域与通道域注意力机制,对特征进行混合空间与通道的混合注意。在特征提取器的输出层融入结合空洞卷积与通道注意力机制的网络层,在不损失空间分辨率的情况下扩大感受野,根据不同感受野提取特征的通道权重后进行融合,得到全局感受野的关键通道特征;引入特征金字塔结构构建特征提取器,提取高分辨率的特征图,大幅提升网络的检测性能。运用基于二阶段的区域生成网络,回归定位更准确的3维目标框。结果 KITTI（A project of Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago）数据集中的实验结果表明,在物体被遮挡的程度由轻到高时,对测试集中的car类别,3维目标检测框的平均精度AP_3D值分别为83.45%、74.29%、67.92%,鸟瞰视角2维目标检测框的平均精度AP_BEV值分别为89.61%、87.05%、79.69%; 对pedestrian和cyclist 类别,AP_3D和AP_BEV值同样比其他方法的检测结果有一定优势。结论 本文提出的3维目标检测网络,一定程度上解决了3维检测任务中卷积神经网络提取的特征缺乏视觉注意力的问题,从而使3维目标检测更有效地运用于室外自动驾驶。     

基于3D高效率视频编码背景的信息隐藏算法

针对秘密信息在公开网络传输的安全性和容量性问题,提出了一种基于高效率视频编码（HEVC）背景的信息隐藏算法。利用HEVC中多视点图像的背景图作为载体,首先对其进行lαβ分解得到三个灰度分量子图,然后选取α和β分量子图作离散余弦变换,最后将秘密信息重复嵌入到载体区域中。隐藏区域选取能量权重较低的α及β分量子图的离散余弦变换中频系数区域,且进行重复嵌入,使得算法具有良好的不可见性和鲁棒性。实验结果表明,所提算法比基于帧间和帧内算法不可见性分别提高16.1%和11.4%,鲁棒性分别提高55.5%和20.2%。     

Incremental Linear Discriminant Analysis Dimensionality Reduction and 3D Dynamic Hierarchical Clustering WSNs

Optimizing the sensor energy is one of the most important concern in Three-Dimensional (3D) Wireless Sensor Networks (WSNs). An improved dynamic hierarchical clustering has been used in previous works that computes optimum clusters count and thus, the total consumption of energy is optimal. However, the computational complexity will be increased due to data dimension, and this leads to increase in delay in network data transmission and reception. For solving the above-mentioned issues, an efficient dimensionality reduction model based on Incremental Linear Discriminant Analysis (ILDA) is proposed for 3D hierarchical clustering WSNs. The major objective of the proposed work is to design an efficient dimensionality reduction and energy efficient clustering algorithm in 3D hierarchical clustering WSNs. This ILDA approach consists of four major steps such as data dimension reduction, distance similarity index introduction, double cluster head technique and node dormancy approach. This protocol differs from normal hierarchical routing protocols in formulating the Cluster Head (CH) selection technique. According to node’s position and residual energy, optimal cluster-head function is generated, and every CH is elected by this formulation. For a 3D spherical structure, under the same network condition, the performance of the proposed ILDA with Improved Dynamic Hierarchical Clustering (IDHC) is compared with Distributed Energy-Efficient Clustering (DEEC), Hybrid Energy Efficient Distributed (HEED) and Stable Election Protocol (SEP) techniques. It is observed that the proposed ILDA based IDHC approach provides better results with respect to Throughput, network residual energy, network lifetime and first node death round.