基于结构化约束的多视角人体检测方法

针对单视角下信息量不足以及多视角不同视角间信息关联困难的问题,提出了基于结构化约束的多视角人体检测方法.首先通过基于块的人体检测模型获取人体局部块信息;然后采用空间仿射变换将不同视角下重叠区域通过变换矩阵的映射关系关联起来;最后针对仿射变换后的区域因遮挡或者存在多目标导致多视角目标关联困难的问题,利用人体局部显著块间的结构化约束为多视角目标匹配构造最大后验概率模型,通过最优求解获取多视角目标匹配结果.实验结果表明,该方法能够利用多视角信息来有效弥补单视角下人体检测中出现的遮挡问题,显著提高了人体检测效果.

特征直连与结构化约束的多视图子空间聚类

多视图子空间聚类作为处理多视图数据的聚类算法,其目的在于学习到一个共识的子空间后用于聚类。但是,现存的多视图子空间聚类算法只是将目标放在了原有的多个视图上,忽略了通过特征直连得到的数据。提出的FSMC算法使原有的多个视图与特征直连视图相互学习,通过误差重构和结构化约束子空间得到一个更加合适的子空间表示,同时还考虑了多视图与特征直连视图的权重关系。最后,在4个基准数据集上进行实验,验证了算法的有效性。

面向视觉着陆的高精度结构化约束特征点研究

为提高视觉着陆过程中无人机的相对定位精度,选取视觉图像中的直线交点作为结构化约束特征点,设计了基于梯度一致性的边缘检测算法,并结合Shi-Tomasi角点检测算法进行结构化约束特征点的粗定位。对LSD直线检测算法进行改进并设计了亚像素角点定位精度改进算法,在结构化约束特征点粗定位的基础上,将其精度提高到亚像素级。基于实际场景中固有约束的结构化约束特征点具有鲁棒性、旋转和尺度不变性,抗干扰能力更强,其高精度定位有利于提高视觉着陆相对定位的精度与可靠性。

利用结构化和一致性约束的稀疏表示模型进行红外和可见光图像融合

红外和可见光图像融合作为图像融合技术中一个重要组成部分,被广泛应用于军事、工业和生活领域。它能够集成两种模态图像的互补信息,融合成一幅信息丰富、质量较好的图像,不仅能够突出目标信息,还能够保持源有图像的纹理信息和一些显著性的细节。本文提出一种新的红外和可见光图像融合方法,在鲁棒稀疏表示模型的基础上增加了结构化稀疏约束,同时结合了图像区域特征相似的一致性约束项,克服现有一些方法所存在的局部模糊和纹理细节丢失等问题,提高了图像融合的精度。本文主要构建了结构化稀疏表示与一致性约束模型,将其应用到红外和可见光图像融合中并进行了求解,将源图像分解为背景信息和显著性信息,再对背景和显著性信息分别设计融合规则,最后利用字典进行重构,获得红外和可见光融合后的图像。实验结果表明,本文提出的融合算法优于现有的一些多聚焦图像融合算法。

考虑参数空间变异结构的结构化交叉约束随机场模拟方法研究

斜坡岩土体抗剪强度参数的空间变异性具有一定的结构性。为研究岩土体参数空间变异结构对边坡失效概率的影响,依据变异函数的内涵推导出变程与相关距离的数学变换关系,并在此基础上提出了结构化交叉约束随机场模拟方法,用以模拟具有互相关性的参数随机场。建立了结构化交叉约束随机场计算模型,研究不同空间变异结构的抗剪强度参数对边坡失效概率的影响。研究结果表明:结构化交叉约束随机场可用于生成模拟具有复杂各向异性空间变异结构的参数随机场,由于考虑了随机偏差、条件数据和空间变异结构,能较为真实地反映地层实际参数,数据波动较条件参数插值场小。可靠性分析结果表明:不考虑抗剪强度参数空间结构分析易高估边坡的失效概率;考虑c′和φ′互相关性时,失效概率随着相关系数的增加而增加,当参数间呈负相关性时更容易高估边坡的失效概率。

针对混合污染的结构化鲁棒低秩恢复算法在人脸识别中的应用

传统的低秩恢复算法在识别有混合污染的人脸图像时,通常只对污染部分进行一种类型的约束,并不能很好地恢复出干净的样本。针对这种情况,提出了结构化鲁棒低秩恢复算法(structured and robust low-rank recovery for mixed contamination,SRLRR)。SRLRR算法利用对二维误差图像的低秩约束移除样本中的连续污染部分,同时利用稀疏约束分离样本中服从拉普拉斯分布的噪声。另外,为了学习到更具有鉴别性的低秩表示,该算法对表示系数进行了块对角结构化约束。在三个常用数据库上的实验证明了SRLRR算法的有效性和鲁棒性。

Yielding and fracture behaviors of coarse-grain/ultrafine-grain heterogeneous-structured copper with transitional interface

Heterogeneous-structured Cu samples composed of coarse-grained(CG) and ultrafine-grained(UFG) domains with a transitional interface were fabricated by friction stir processing, in order to investigate the effect of interface constraint on the yielding and fracture behaviors. Tensile test revealed that the synergetic strengthening induced by elastic/plastic interaction between incompatible domains increases with increasing the area of constraint interface. The strain distribution near interface and the fracture morphology were characterized using digital image correlation technique and scanning electron microscopy, respectively. Fracture dimples preferentially formed at the interface, possibly due to extremely high triaxial stress and strain accumulation near the interface. Surprisingly, the CG domain was fractured by pure shear instead of the expected voids growth caused by tensile stress.

A hybrid cuckoo search algorithm with feasibility-based rule for constrained structural optimization

Constrained optimization problems are very important as they are encountered in many science and engineering applications.As a novel evolutionary computation technique,cuckoo search(CS) algorithm has attracted much attention and wide applications,owing to its easy implementation and quick convergence.A hybrid cuckoo pattern search algorithm(HCPS) with feasibility-based rule is proposed for solving constrained numerical and engineering design optimization problems.This algorithm can combine the stochastic exploration of the cuckoo search algorithm and the exploitation capability of the pattern search method.Simulation and comparisons based on several well-known benchmark test functions and structural design optimization problems demonstrate the effectiveness,efficiency and robustness of the proposed HCPS algorithm.

Global signatures of protein binding on structured RNAs in Saccharomyces cerevisiae

Protein binding is essential to the transport,decay and regulation of almost all RNA molecules.However,the structural preference of protein binding on RNAs and their cellular functions and dynamics upon changing environmental conditions are poorly understood.Here,we integrated various high-throughput data and introduced a computational framework to describe the global interactions between RNA binding proteins(RBPs)and structured RNAs in yeast at single-nucleotide resolution.We found that on average,in terms of percent total lengths,~15%of mRNA untranslated regions(UTRs),~37%of canonical non-coding RNAs(ncRNAs)and^11%of long ncRNAs(lncRNAs)are bound by proteins.The RBP binding sites,in general,tend to occur at single-stranded loops,with evolutionarily conserved signatures,and often facilitate a specific RNA structure conformation in vivo.We found that four nucleotide modifications of tRNA are significantly associated with RBP binding.We also identified various structural motifs bound by RBPs in the UTRs of mRNAs,associated with localization,degradation and stress responses.Moreover,we identified>200 novel lncRNAs bound by RBPs,and about half of them contain conserved secondary structures.We present the first ensemble pattern of RBP binding sites in the structured non-coding regions of a eukaryotic genome,emphasizing their structural context and cellular functions.