地铁站场景下的可行驶区域检测

在室内环境下的机器人视觉导航任务中,可行驶区域检测是不可或缺的一部分,这是保证自动驾驶任务实现的基础.目前较多的解决方法是对数据集中出现过的障碍物进行识别来检测可行驶区域,缺乏灵活性,因此本文提出了一种针对地铁站等室内平坦地面的可行驶区域检测方法,提高实用性.本文采用经典的MobileNetV3网络对采集到的前方图像进行分类,判断是否为地面区域.由于室内地面的地标、箭头等贴纸的影响,因此需要对非地面区域进一步判断,与常规的立体障碍物进行区分.本文利用连续帧之间的特征点匹配获得相机移动距离,并利用直线拟合计算斜率的方法达到区分立体障碍物与平面地标的目的.实验表明,本文提出的方法能较好地检测机器人前方可行驶区域,具有较高的实用价值.     

PointPCA:一种三维空间点云的特征提取算法

点云是一个庞大的点的集合而且拥有重要的几何结构。由于其庞大的数据量,不可避免地就会在某些区域内出现一些相似点,这就使得在进行特征提取时提取到一些重复的信息,造成计算冗余,降低训练的准确率。针对上述问题,提出了一种新的神经网络——PointPCA,可以有效地解决上述问题;在PointPCA中,总共分为三个模块：a）采样模块：提出了一种average point samping（APS）采样方法,可以有效地规避一些相似的点,得到一组近似代表这组点云的新的点集;b）特征提取模块：采用分组中的思想,对这组新的点的集合进行多尺度空间特征提取;c）拼接模块：将每一尺度提取的特征向量拼接到一起组合为一个特征向量。经过实验表明,PointPCA比PointNet在准确率方面提升了4.6%,比PointNet++提升了1.1%;而且在mIoU评估测试中也有不错的效果。     

一种面向低采样率的点云数据处理网络

在直接处理点云的三维神经网络中,采样阶段实现了对原始点云中关键点的筛选,对于整个网络的性能及网络的抗噪能力具有重要作用。目前主流的最远点采样（FPS）方法在处理大规模3D点云数据时计算量大且耗时,并且低采样率时经过FPS采样后模型性能下降明显。针对这两个问题,提出一种面向低采样率的点云数据处理网络AS-Net。设计一个新的采样模块代替原backbone中的FPS,其由两个Layer组成,每个Layer基于长短期记忆网络获取原始点云与采样点云之间的联系权重,从而高效提取关键信息,去除冗余信息。在此基础上,利用注意力机制选择特征值较高的原始点云作为采样点,采样点作为后序任务的关键点输入到网络,进一步提高网络模型性能。基于ModelNet40数据集的实验结果表明,在低采样率条件下,AS-Net仍可达到81.6%的分类准确率,与使用FPS作为采样方法的网络模型相比提高52.7%。此外,其对噪声干扰具有很强的鲁棒性,对于大场景的分割时间效率优于同类采样方法。     

Reliability analysis on systems with self-healing and self-repairing under different environments and shock models

Inspired by the process of self-healing of biological damage, high technology materials with self-healing and self-repairing mechanisms have been developed for high reliability and long lifetime. Therefore, the reliability modeling on intelligent systems with healing performance has become a research hotspot. Based on the diversity of healing mechanisms, this paper proposes a two-phase reliability model method on self-healing and self-repairing systems. Impacts of environments, shock loads, self-healing, and self-repairing mechanisms are taken into account in this novel model. Besides, system lifetime and some reliability indexes under two shock models are derived, respectively. Moreover, Monte Carlo simulations are conducted to verify the accuracy of reliability under two models. Finally, an engineering case of metallized film capacitor is provided to illustrate the effectiveness and applicability of the proposed models by comparing numerical results and simulation results.     

从测角后交最佳点位规律看规范规定之不足

本文分析了文献[1]所总结的测角后交两类公式在取得最佳点位时的规律,指出当前规范对此问题规定之不足.     

Influence of ions on the critical behavior of a binary mixture near the consolute point

The lower critical point of stratification of a 3-methylpyridine (MP)+heavy water (D₂O) mixture in the presence of Na⁺ and CI^– ions has been studied by the Toepler shadow method. Addition of 0.3% ions lowered the critical temperature and reduced the equilibration time and the gradient of the refractive index (compressibillity). The analysis of the form of the near-critical isotherm demonstrated the ionic mixture to be described by the index = 3.05 ± 0.15. which corresponds to the classical mean-field theory. The results obtained provide evidence that even small admixtures of charged particles result in a substantial suppression of fluctuations near the critical point by the long-range Coulomb interaction.Paper presented at the Twelfth Symposium on Thermophysical Properties, June 19–24, 1994, Boulder, Colorado, U.S.A.     

JSG-8型束管火灾监测系统井下布置方案探讨

介绍了JSG-8型束管火灾监测系统的工作原理及结构;针对工作面、采空区煤自燃指标气体采集困难的问题,详细阐述了该系统束管测点在工作面及采空区的布置方案,并指出在工作面进行束管测点布置时需考虑对工作面生产的影响程度。     

基于行扫描的炮弹引信铆点特征提取

在炮弹引信铆点机器视觉识别中,有效消除图像中各种干扰、正确提取铆点特征是关键,解决好铆点特征的准确提取问题,是炮弹引信安全破除铆点的前提条件;根据铆点的结构特点,分析了产生各种干扰的原因及特点,在高斯滤波,阀值分割、形态学运算等基本运算的基础上,提出了一种基于行扫描的铆点特征提取方法;测试结果表明,提出的方法可行,铆点正确识别率达到90%以上,具有高效性和很好的鲁棒性的铆点特征提取,使得炮弹铆点的准确破除成为可能.     

尖端导体电晕放电试验研究

电晕放电是高压电力工程及航空航天领域研究的重点及难点问题;为研究尖端导体电晕放电的辐射特点,设计了电晕放电试验模拟系统;利用测试系统及天线在微波暗室内进行了不同测试距离、不同放电电压及不同尖端导体长度条件下的试验;通过对试验数据的分析可知,电晕放电辐射信号呈衰减振荡模式;电晕放电脉冲重复率随着充电电压的增大而增加;天线接收的辐射信号能量与测试距离成反比;随着充电电压的增加,信号辐射功率呈增加趋势;试验研究结果为输变电工程、航空航天装备的电磁环境分析提供参考。     

Critical point analysis using domain lifting for fast geometry queries

In this paper, a general scheme for solving coherent geometric queries on freeform geometry is presented and demonstrated on a variety of problems common in geometric modeling. The underlying strategy of the approach is to lift the domain of the problem into a higher-dimensional space to enable analysis on the continuum of all possible configurations of the geometry. This higher-dimensional space supports analysis of changes to solution topology by solving for critical points using a B-spline-based constraint solver. The critical points are then used to guide fast, local methods to robustly update repeated queries. This approach effectively combines the speed of local updates with the robustness of global search solutions. The effectiveness of the domain lifting scheme (DLS) is demonstrated on several geometric computations, including accurately generating offset curves and finding minimum distances. Our approach requires a preprocessing step that computes the critical points, but once the topology is analyzed, an arbitrary number of geometry queries can be solved using fast local methods. Experimental results show that the approach solves for several hundred minimum distance computations between planar curves in one second and results in a hundredfold speedup for trimming self-intersections in offset curves.