基于人体姿态估计的分心驾驶行为检测

为弥补现有驾驶特征提取方法的不足,提高分心驾驶行为检测的准确性和鲁棒性,将2D/3D人体姿态估计应用于驾驶人行为检测,提出一种适用于驾驶舱环境下的驾驶特征提取方法。首先通过将2D姿态估计网络Simple Baseline和分类网络ResNet进行融合,构建基于2D姿态估计的分心驾驶行为检测模型,并在分心驾驶数据集State Farm上分析不同数据增强方法、不同超参数、不同分类网络对模型性能的影响。其次,融合3D密集姿态估计网络DensePose与分类网络ResNet,构建基于3D姿态估计的分心驾驶行为检测模型。接着,在State Farm数据集上,针对模型的实时性和泛化能力,对比分析基于原始图像和基于2D/3D姿态的分心驾驶行为检测模型。最后,针对效果更优的基于2D姿态估计的分心驾驶行为检测模型,在分心驾驶数据集State Farm上,对使用不同姿态估计算法和分类网络的分心驾驶行为检测模型做了交叉试验,对比分析4个不同检测模型的优缺点。进一步地,将基于2D姿态估计的分心驾驶行为检测模型应用于实际采集的驾驶图片,对模型的泛化能力和有效性进行了测试验证。研究结果表明:与基于原始图像的检测模型相比,基于2D和3D姿态的检测模型都能显著提高分心驾驶行为的检测准确率;基于3D姿态的检测模型在检测精度方面略优,但基于2D姿态的检测实时性更好,检测效率是基于3D姿态检测的4倍;在驾驶舱单一环境下,基于2D姿态估计的分心驾驶行为检测模型能够满足分心驾驶行为检测的需求,在分心驾驶行为检测方面具有重要应用价值。     

基于改进BOMP算法的水声信道估计

近年来,水声信道估计主要是基于稀疏模型展开.水声介质的非均匀性等使声线以簇的形式传播,导致水声信道展现出块结构稀疏特性.本文针对信道的块结构稀疏特性,在OFDM通信系统中,提出使用改进的BOMP算法进行水声信道估计.BOMP算法一次筛选1个最大相关块,改进的算法一次挑选t个非零块,算法重构时间将降低t倍.仿真结果表明:改进的BOMP算法误码率和重构时间要优于传统的LS、基于压缩感知的OMP算法;在不降低BOMP算法重构精度的前提下,将重构时间降低t倍.     

基于视觉的集装箱海上并靠吊装位姿识别

舰船在远海进行并靠补给时受波浪影响两船之间会产生无规律的相对运动，给补给集装箱吊装工作带来了很大挑战，急需一种能够在较高海况下辅助人工的自动化吊装系统。针对这一问题，本文提出了一种基于单目视觉和ArUco标志点相结合的集装箱位姿检测方法。首先，介绍了带有视觉系统的并联吊装系统主要组成部分，简要分析了控制流程及方法；然后，根据实际集装箱码放流程，设计了适用的ArUco标志点布放和检测方案，介绍了相机标定、ArUco标志点检测及集装箱位姿估计方法；最后，搭建模拟试验平台测试视觉位姿识别的精度、速度和不同光照条件下的鲁棒性。试验结果表明，在普通工控PC平台上，最低识别速度在15Hz以上，姿态角平均误差为 ，位移量平均误差为 ，可以满足并靠吊装要求。     

A simplified life estimation method for the spherical hull of deep manned submersibles

Creep and fatigue are involved in the loading of deep manned submersible, which is a rather complex variable amplitude pattern. Dwell effects resulting in lower life than pure fatigue are observed in the prior experimental research while no proper prediction methods are available to explain the phenomenon. Recently, the authors proposed a modified crack growth rate model to explain the creep effect under the cyclic loading which is validated for the crack growth of some titanium alloys under cyclic creep loading, however, its application is restricted to the condition where lots of parameters have to be determined based on many experiments and it is not very convenient in the material selection and primary design stage of a component. In this paper, a simplified prediction model for the load pattern of constant amplitude cyclic loading is proposed aiming at one of the most applicable materials for the pressure hull of submersibles, Ti–6Al–4V ELI. The method can be easily used to estimate the life of the pressure hull based on two series of basic mechanical properties of the material and validated by the modified crack growth rate model with parameters determined by large amounts of experimental data.     

基于电子支付数据的公交车厢满载率实时估算方法

针对公交监测和调度中要求实时掌握车厢满载率，以及“一票制”无法获取乘客下车信息等问题，构建基于数据驱动的组合模型，在乘客上车时即推断其出行OD站点，进而融合多源数据实现车厢满载率的实时估算。提出以K近邻算法为组合模型的核心，针对K近邻推断率过低等问题，研究在更大空间维度分析乘客出行规律并推断下车站点的方法，有效提升历史数据的利用率和下车站点的推断率；此外，针对偶发型乘客缺少历史规律数据的情况，充分利用站点下车客流量先验概率随机分配，实现电子支付乘客OD的全样本推断。利用跟车调查法对不同线路、不同班次的车厢拥挤度进行验证。结果表明，模型计算结果与实际结果相符，能够反映出不同线路、不同站段之间的车厢拥挤水平变化。     

基于实测应力的动车组制动盘盘毂服役特性

针对仿真或仅考虑紧急制动状态下动车组制动盘盘毂安全性分析中存在的不足,基于盘毂应力在线测试,分析动车组高速运行和不同制动方式下盘毂的频谱特性和成分特性;考虑应力集中,根据静力等效原则进行毂齿根部的应力线性化,分析不同制动方式对盘毂疲劳损伤的影响;采用指数模型拟合和核密度估计相结合的方法,推理97.5%置信度下的盘毂实测应力谱,并考虑车轮镟修前后盘毂损伤演化和材料强度退化,评估盘毂服役安全性。结果表明:盘毂载荷振动频率主要分布在0~71 Hz和341~680 Hz频带,车轮镟修可有效降低341~518 Hz频带内的载荷振动、消除518~680 Hz频带内的载荷振动;动车组高速运行和车轮状态不良是造成盘毂损伤的主要原因;若按盘毂服役寿命为1500万km计算,盘毂疲劳薄弱区的等效应力为37.4 MPa,累积损伤为0.61,该结果可为盘毂的结构设计和检修维护策略制定提供依据。     

考虑动力学模型系统误差补偿的智能车GNSS/IMU组合定位算法

为了解决智能车动态组合定位过程中，因动力学模型与实际模型之间存在偏差导致滤波精度下降的问题，针对智能车全球导航卫星系统(GNSS)/惯性测量单元(IMU)组合定位系统，结合非线性预测滤波(NPF)和自适应滤波的优点，提出了一种考虑动力学模型系统误差实时估计和补偿的自适应非线性预测滤波(ANPF)算法。首先，根据NPF算法原理，通过最小化预测观测残差与系统误差的加权平方和，估计动力学模型系统误差；其次，结合自适应滤波原理，利用状态预测残差向量构造自适应因子，设计了一种自适应扩展卡尔曼滤波(AEKF)算法，用于估计系统状态向量，并通过自适应因子抑制动力学模型系统误差和线性化误差对系统状态估计精度的影响，克服NPF对系统状态估计精度有限的缺陷；再次，对动力学模型系统误差的估计误差和由动力学模型系统误差引起的系统噪声的等效协方差阵进行了分析和推导，以补偿动力学模型系统误差对系统状态估计的影响；最后，通过车载GNSS/IMU组合定位系统试验，从算法精度、鲁棒性和实时性方面对提出的算法和其他滤波算法的性能进行了验证和对比分析。研究结果表明:提出的自适应算法继承了NPF算法简易性和高实时性的优点，同时克服了NPF算法估计精度有限的缺陷，具有较好的滤波解算精度，水平定位精度小于1.0 m，算法单次平均执行时间约为0.013 9 ms，在精度和实时性的平衡方面显著优于其他滤波方法。     

A distribution-fitting-free approach to calculating travel time reliability ratio

Empirical studies have revealed that travel time variability (TTV) can significantly affect travelers’ behaviors and planners’ cost-benefit assessment of transportation projects. It is therefore important to systematically quantify the value of TTV (VTTV) and its impact. Recently, Fosgerau’s valuation method makes this quantification possible by converting the value of travel time (VTT) and the VTTV into monetary unit. Travel time reliability ratio (TTRR), defined as a ratio of the VTTV to the VTT, is a key parameter in Fosgerau’s valuation method. Calculating TTRR involves an integral of the inverse cumulative distribution function (CDF) of the standardized travel time distribution (STTD), i.e., the mean lateness factor. Using a well-fitted STTD is a straightforward way to calculate TTRR. However, it will encounter the following challenges: (1) determination of a well-fitted STTD; (2) non-existence of an algebraic expression for the CDF and its inverse CDF; and (3) lack of a closed-form expression to efficiently calculate TTRR. To circumvent the above issues, this paper proposes a distribution-fitting-free analytical approach based on the Cornish-Fisher expansion as an alternative way to calculate TTRR without the need to fit the whole CDF. The validity domain is rigorously derived for guaranteeing the accuracy of the proposed method. Realistic travel time datasets that cover 17 links are used to systematically explore the feature and accuracy of the proposed method in estimating TTRR. The comparative results demonstrate that the proposed method can efficiently and effectively estimate TTRR. When travel time datasets satisfy the validity domain, the proposed method outperforms the distribution fitting method in estimating TTRR.     

考虑人口效益的城市路网OD 矩阵推算模型

交通分布数据获取难度大、成本高，是制约交通工程师开展交通预测的主要因素. 为提高效率、降低工程实践中的预测成本，对城市路网OD矩阵可预测性展开研究. 描述3 种交通分布推算模型并以广州为例，给出交通流概率分布及发生量概率分布. 通过回归分析及残差分析探究发生量与人口、经济的关系，提出考虑人口效益的靶向双联模型(称为TDM模型). 根据误差分析验证4 种模型准确度，并将模型应用于深圳市. 结果表明：交通流及发生量概率分布具有高度异质性，遵循Zipf 定律；发生量与人口、经济相关性强，拟合优度为0.87； TDM模型精度略低于重力模型，但高于其他两种模型，且在深圳市推算效果良好. 综合预测精度、成本和效率，TDM模型更适合预测城市道路交通分布.     

Review of optimal sensor location models for travel time estimation

The problem of optimally locating fixed sensors on a traffic network infrastructure has been object of growing interest in the past few years. Sensor location decisions models differ from each other according to the type of sensors that are to be located and the objective that one would like to optimize. This paper surveys the existing contributions in the literature related to the problem of locating fixed sensors on the network to estimate travel times. The review consists of two parts: the first part reviews the methodological approaches for the optimal location of counting sensors on a freeway for travel time estimation; the second part focuses on the results related to the optimal location of Automatic Vehicle Identification (AVI) readers on the links of a network to get travel time information.