这里写目录标题

  • 0.Abstract:
    • 0.1逐句翻译
    • 0.2总结
  • 1. Introduction
    • 1.1逐句翻译
      • 第一段(当前的定位应用很发达,但是室内定位是一个缺口)
      • 第二段(当前的主要方式-基于指纹类的)
      • 第三段(这里介绍行人航迹推算的内容,会随着时间变飘,有比较好的算法,但是那个是绑在脚上的)
      • 第四段(介绍其他的人的个钟头方式,这些方式都多少还有些问题)
      • 第五段(PDR的一个关键是航向,有些论文提出使用卡尔曼滤波来进行修正)
      • 第六段(介绍本文提出的新方法)
      • 第七段(介绍本文接下来写的什么东西)
    • 1.2总结
  • 2. Materials and Methods
    • 2.1. Proposed System Scheme(提出的解决方案)
      • 2.1.1逐句翻译
        • 第一段(介绍传统的PDR和本文的坐标系规定)
        • 第二部分(介绍传统方法的航向、位置、速度获得的方法)
        • 第三段(简单介绍本文的方法)
      • 2.1.2总结
    • 2.2. Step Detection Based On State Transition
      • 第一部分(叙述怎么处理数据的)
      • 第二部分(具体描述怎么进行状态转化的)
      • 第三部分

0.Abstract:

0.1逐句翻译

With the rapid development of smartphone technology, pedestrian navigation based on built-in inertial sensors in smartphones shows great application prospects.
随着智能手机技术的快速发展,基于智能手机内置惯性传感器的行人导航显示出了巨大的应用前景。(用手机的多了,用手机定位的场景自然就多了)

Currently, most smartphone-based pedestrian dead reckoning (PDR) algorithms normally require a user to hold the phone in a fixed mode and, thus, need to correct the gyroscope heading with inputs from other sensors, which restricts the viability of pedestrian navigation significantly.
目前,大多数基于智能手机的行人航位推算(PDR)算法通常要求用户手持手机处于固定模式,因此需要使用其他传感器的输入来修正陀螺仪的航向,这极大地限制了行人导航的可行性。

In this paper, in order to improve the accuracy of the traditional step detection and step length estimation method for different users, a state transition-based step detection method and a step length estimation method using a neural network are proposed.
为了提高传统的步长检测和步长估计方法对不同用户的精度,提出了一种基于状态转移的步长检测方法和一种基于神经网络的步长估计方法。

In order to decrease the heading errors and inertial sensor errors in multi-mode system, a multi-mode intelligent recognition method based on a neural network was constructed.
为了减小多模态系统的航向误差和惯性传感器误差,提出了一种基于神经网络的多模态智能识别方法。
(这个文章使用神经网络识别了一种持有姿态)

On this basis, we propose a heading correction method based on zero angular velocity and an overall correction method based on lateral velocity limitation (LV).
在此基础上,提出了一种基于零角速度的航向校正方法和一种基于横向速度限制(LV)的整体校正方法。

Experimental results show that the maximum positioning errors obtained by the proposed algorithm are about 0.9% of the total path length.
实验结果表明,该算法获得的最大定位误差约为总路径长度的0.9%。

The proposed novel PDR algorithm dramatically enhances the user experience and, thus, has high value in real applications.
该算法显著提高了用户体验,具有较高的实际应用价值。

0.2总结

大约就是提出了下面的组成部分:

  • 1.提出了一种状态机的步长估计和一种神经网络的步长估计
  • 2.一种基于神经网络的持有姿态识别
  • 3.一种基于角度限制的航向修正。

然后取得了很好的效果

1. Introduction

1.1逐句翻译

第一段(当前的定位应用很发达,但是室内定位是一个缺口)

With the development of society, location-based service became part of people’s lives.
随着社会的发展,基于位置的服务成为人们生活的一部分。

A pedestrian navigation and positioning service system mainly depends on a global positioning system (GPS) in an outdoor environment [1,2].
行人导航定位服务系统主要依赖于室外环境下的全球定位系统(GPS)[1,2]。(就是室外可以使用GPS,但是室内就没有这个东西了)

For indoor applications in, e.g., airports, train stations, underground garages, and shopping malls, GPS often becomes unavailable because of signal occlusion, which would hinder the application of location-based services in these areas.
对于机场、火车站、地下车库、商场等室内应用,GPS常常因为信号遮挡而无法使用,这将阻碍基于位置的服务在这些区域的应用。

Therefore, pedestrian indoor navigation technology is critical to ensure the success of location-based services.
因此,行人室内导航技术是确保位置服务成功的关键。

第二段(当前的主要方式-基于指纹类的)

Currently, pedestrian indoor navigation and positioning can be achieved using two types of methods.

The first type of method is based on wireless technologies, e.g., WiFi [3–5], ultra-wideband (UWB) [6,7], visual sensors [8], radio frequency identification (RFID) [9], ibeacon [10], Bluetooth, and/or ZigBee, with a multi-source information fusion technique [11] to obtain pedestrian location information.
第一种方法是基于无线技术,如WiFi[3-5]、超宽带(UWB)[6,7]、视觉传感器[8]、射频识别(RFID)[9]、ibeacon[10]、蓝牙和/或ZigBee等,采用多源信息融合技术[11]获取行人位置信息。
(这些大约就是指纹类的方法来完成这个任务)

For the first type of method, its location errors do not accumulate over time. However, these methods need a significant cost to deploy wireless devices as beacons before navigation and are infeasible for the unknown environment.
对于第一种方法,它的定位误差不会随时间累积。然而,这些方法需要在导航前部署无线设备作为信标,并且在未知环境下是不可行的。
(这个东西是比较准的就是关键是得提前部署这些东西,一定程度上限制了使用)

第三段(这里介绍行人航迹推算的内容,会随着时间变飘,有比较好的算法,但是那个是绑在脚上的)

In these methods, an inertial sensor is fixed on the foot [13,14], waist [15,16], leg, or shoulder of a user for data acquisition, in order to determine the position of pedestrian using the number and the size of walking steps and the heading angle [17].
在这些方法中,将惯性传感器固定在用户的脚[13,14]、腰[15,16]、腿或肩膀上进行数据采集,利用行人行走的步数、步数大小和行走的方向角[17]来确定行人的位置。

These methods belong to autonomous positioning technologies and have a high positioning accuracy in a short-time period.
这些方法属于自主定位技术,在短时间内具有较高的定位精度。
(推久了就开始飘了)

However, their positioning performance is drastically affected by the accumulative errors of iterative calculation.
但其定位性能受迭代计算的累积误差影响较大。

Therefore, many researchers are committed to solving the accumulative errors in the traditional PDR algorithm [18,19].
因此,许多研究者致力于解决传统PDR算法中的累积误差问题[18,19]。

Zhou et al. [19] proposed a PDR algorithm with low computational complexity using a foot-mounted inertial measurement unit (IMU) and estimating orientation of the foot based on extended Kalman filter (EKF), which can correct the accumulative errors of PDR.
Zhou et al.[19]提出了一种计算复杂度较低的PDR算法,该算法利用足部惯性测量单元(IMU)和基于扩展卡尔曼滤波(EKF)的足部方位估计,可以修正PDR的累积误差。

Nonetheless, it can only be applied for foot-mounted navigation and, therefore, has poor viability for different users.
然而,它只能应用于脚踏导航,因此,对于不同的用户来说,可行性很差。(但是这种比较准的算法是绑在脚上的)

第四段(介绍其他的人的个钟头方式,这些方式都多少还有些问题)

The MEMS-based algorithm brings inconvenience to practical application due to the extrainstallation of inertial sensors.
MEMS-based的算法由于惯性传感器的额外安装给实际应用带来了不便。

In allusion to the inconvenience of pedestrian positioning technology based on fixed inertial sensors, smartphone-based pedestrian navigation gradually became a research hotspot for experts.
针对基于固定惯性传感器的行人定位技术带来的不便,基于智能手机的行人导航逐渐成为专家们的研究热点。

Bilke et al. [20] proposed a locating system based on recognizing geomagnetic field disturbances and ambient light.
Bilke et al.[20]提出了一种基于地磁场扰动和环境光识别的定位系统。

Li [21] combined Bluetooth and an inertial navigation system to implement a set of navigation systems based on a smartphone.
Li[21]结合蓝牙和惯性导航系统,实现了一套基于智能手机的导航系统。

Kang and Han [22] designed a smartphone-based algorithm, named smartPDR, but users had to reduce the sway of the smartphone in experiments.
Kang和Han[22]设计了一种基于智能手机的算法,名为smartPDR,但在实验中,用户必须减少智能手机的摇摆。

Wang et al. [23] presented a PDR approach based on motion mode recognition using a smartphone.
Wang et al.[23]提出了一种基于智能手机运动模式识别的PDR方法。

The motion mode recognition was achieved using a support vector machine (SVM) and a decision tree (DT), which thereby increased the complexity of the algorithm.
运动模式识别采用支持向量机(SVM)和决策树(DT)实现,增加了算法的复杂性。(如果增加运动模式识别,至少需要使用到某种机器学习模型,所以复杂度会有所上升)

Zeng et al. [24] proposed an autonomous inertial heading correction algorithm based on the Kalman filter and the different usage modes of the smartphone, e.g., normal mode, landscape mode, or call mode.
Zeng等人[24]提出了一种基于卡尔曼滤波的自主惯性航向校正算法,该算法结合智能手机的不同使用模式,如正常模式、横向模式、通话模式。
(大约就是用卡尔曼滤波进行误差修正)

The usage mode could be determined based on the gravity-assisted (GA) method. However, in References [22–24], step detection methods were both based on peak detection, which would introduce detection errors due to the step detection errors.
利用重力辅助(GA)方法可以确定使用模式。但是在文献[22-24]中,阶跃检测方法都是基于峰值检测的,由于阶跃检测误差会引入检测误差。

Moreover, the step length estimation method was based on a nonlinear model or constant which would cause position errors for different users.
此外,步长估计方法是基于非线性模型或常数,对不同的用户会产生位置误差。

Furthermore, mode recognition methods do not have universality, as the accuracy of these methods is affected by the configurations of the threshold.
此外,模式识别方法不具有通用性,因为这些方法的准确性受到阈值配置的影响。

第五段(PDR的一个关键是航向,有些论文提出使用卡尔曼滤波来进行修正)

The heading error is the one of the main error sources in PDR.
航向误差是PDR的主要误差来源之一。

In References [20–22], smartphone-based navigation algorithms required users to hold a smartphone in a fixed mode as long as possible.
在文献[20-22]中,基于智能手机的导航算法要求用户尽可能长时间地保持智能手机的固定模式。
(PDR的一个关键就是获得一个准确的航向,所以这就要求大家使用一个固定的状态拿着手机,来获得一个准确的航向)

However, it is difficult for users to keep a fixed mode during the movement process, which leads to significant heading errors.
但在移动过程中,用户难以保持固定的模式,导致了较大的航向误差。

Wang et al. [23] proposed a principal component analysis (PCA)-based method with global accelerations (PCA-GA) to infer pedestrian headings.
Wang et al.[23]提出了一种基于主成分分析(PCA)的全局加速度(PCA- ga)方法来推断行人标题。

However, one set of experiments represented only one phone pose, which is not applicable to multi-mode changes of a smartphone.
然而,有一组实验只代表了一种手机姿势,这并不适用于智能手机的多模式变化。

Although Reference [24] proposed using a Kalman filter to correct the heading errors in a multi-mode application, heading correction values are hard to estimate precisely due to weak constraints and observation
errors.
虽然参考文献[24]提出在多模式应用中使用卡尔曼滤波器来校正航向误差,但由于约束条件和观测误差较弱,难以精确估计航向修正值。

第六段(介绍本文提出的新方法)

In this paper, we propose a novel pedestrian dead reckoning algorithm for multi-mode recognition based on a smartphone.
本文提出了一种基于智能手机的行人航迹推算多模式识别算法。

Firstly, different from the peak detection in References [22–24], state transition was carried out to improve the accuracy of step detection for different phone modes.
首先,与文献[22-24]中的峰值检测方法不同的是,为了提高对不同电话模式的计步器的准确性,我们进行了状态转移。
(为了提升不同手机姿态下步伐检测的准确性本文提出了一种状态转移的计步器)

Secondly, we adopted the neural network to obtain the step length and the smartphone mode without setting thresholds manually, which increased the universality of step length estimation and mode recognition methods.
其次,在不手动设置阈值的情况下,采用神经网络获取步长和智能手机模式,提高了步长估计和模式识别方法的通用性;

The main contributions of the paper are that we design the heading correction method based on zero angular velocity (ZA) and an overall correction method based on lateral velocity limitation (LV) to correct the error of heading estimation and inertial sensors in a smartphone-based PDR.
本文的主要贡献在于设计了基于零角速度(ZA)的航向修正方法和基于横向速度限制(LV)的整体航向修正方法,对基于智能手机的PDR中的航向估计和惯性传感器的误差进行了修正。

Experimental results show that the algorithm proposed in this paper is effective.
实验结果表明,本文提出的算法是有效的。

第七段(介绍本文接下来写的什么东西)

The rest of the paper is organized as follows:

Section 2 presents our proposed algorithm in detail, including the comparison between the traditional PDR algorithm and the proposed novel PDR algorithm, along with the step detection, step length estimation, multi-mode intelligent recognition,and heading correction methods.
第2节详细介绍了我们提出的算法,包括传统PDR算法和新PDR算法的比较,以及步长检测、步长估计、多模式智能识别和航向校正方法。

The experimental results are systematically analyzed in Section 3.
实验结果将在第三部分进行系统分析。

Sections 4 and 5 discuss the system and draw conclusions.
第4节和第5节讨论了系统并得出结论。

1.2总结

这里大约介绍了:

  • 1.随着社会发展定位的应用越来越广泛,但是室内定位仍然存在问题。
  • 2.室内定位存在两种方法一种是指纹的方法,一种是行人航迹推算的方法
  • 3.之前比较准的算法是绑在脚上的,这个绑在脚上限制了应用。
  • 4.介绍其他人的具体方法,反正就是都有不足。
  • 5.航向的误差也是一个急需解决的问题。
  • 6.本文使用的方法。

别管怎样这里大约就是说了之前的两个问题:

  • 1.手机的持有状态是不一样的,所以需要采用多种不同的方式来处理
    这里的处理其实包括两方面:1)怎么识别这些状态 2)怎么应对这些状态。
  • 2.随着时间传感器推算会逐渐变得不准确。

所以本文提出了一个方法:

  • 1.首先使用神经网络获得一个持有姿态。
  • 2.之后根据不同的持有姿态获得不同的计步器和步伐检测。
  • 3.之后使用零角度速度和时序上长时间的速度限制,修正长时间的传感器漂移

2. Materials and Methods

2.1. Proposed System Scheme(提出的解决方案)

2.1.1逐句翻译

第一段(介绍传统的PDR和本文的坐标系规定)

As shown in Figure 1, using an accelerometer and a gyroscope, the traditional scheme of the PDR algorithm estimates the position, velocity, and heading using the results of step detection, step length estimation, and heading estimation.
如图1所示,PDR算法的传统方案使用加速度计和陀螺仪,利用阶跃检测、步长估计和航向估计的结果估计位置、速度和航向。
(大约就是介绍一下之前的PDR都是怎么做的)

In Figure 1, f and ω are the outputs of the accelerometer and gyroscope, respectively, flag indicates the result of step detection, L is the step length estimation, and φ is the heading estimation of the pedestrian.
译文:在图1中,f和ω分别为加速度计和陀螺仪的输出,flag为步长检测结果,L为步长估计,φ为行人的航向估计。

e result of step detection, L is the step length estimation, and Ф is the heading estimation
行人导航涉及人体坐标系和导航坐标系。

The x-axis of the body coordinate system corresponds to the lateral direction of the smartphone, the y-axis corresponds to the forward direction of the smartphone, and the z-axis satisfies the right-hand rule.
身体坐标系的x轴对应智能手机的横向方向,y轴对应智能手机的正向方向,z轴满足右手规则。(也就是右前上载体坐标系)

The navigation coordinate system involves east, north, up (ENU) coordinates.
导航坐标系为东北天。

第二部分(介绍传统方法的航向、位置、速度获得的方法)

The traditional heading estimation method uses a z-axis gyroscope ωzk, denoted by
传统的航向估计方法采用z轴陀螺仪ωzk,表示为

where ϕk and ϕk−1 are the headings at time instants kT and (k − 1)T, respectively. The initial heading is set manually.
其中ϕk和ϕk - 1分别是时刻kT和(k - 1)T的航向。初始航向是手动设置的。
(这里注意heading是定位当中航向的意思)
大致想一下就是这里使用z轴的陀螺仪数据是可以用来判定航向的变化的,以此来推算当前航向的变化。(但是这个东西一切都是建立在手机是平端的情况下的)

The traditional position estimation method is resolved using the step length Li and the heading of the i-th step, denoted by
传统的位置估计方法是用步长Li和第i步的方向来解决的,用

也就是使用方向加上步长得到移动的距离
The velocity estimation method is calculated by
速度估计方法是由

where T represents sampling period, fx,k and fy,k are the lateral acceleration and forward acceleration on the horizontal plane at time instants kT, and fe,k and fn,k are the eastward acceleration and northward acceleration at time instants kT.
式中,T为采样周期,fx、k、fy,k为时刻kT水平面上的横向加速度和正向加速度,fe、k、fn、k为时刻kT的东向加速度和北向加速度。

Then, the eastward velocity and northward velocity are calculated by

(大约就是我们通过载体坐标系加上航向角,其实就是姿态角可以解算出北向和东向的加速度,可以分别的获得北向和东向的速度,这里其实当前这个公式有个什么问题呢就是这个公式当中默认加速度是不变的。所以这就存在误差。)

第三段(简单介绍本文的方法)

The accurate estimations of step, step length, and heading are crucial for a pedestrian navigation algorithm.
准确估计步数、步长和方向是行人导航算法的关键。

Since the accuracy of built-in inertial sensors in a smartphone is at a lower level,the traditional PDR algorithm would have relatively large navigation errors; in particular, the drift of heading can result in a drastic increase in errors.
由于智能手机内置惯性传感器的精度较低,传统的PDR算法存在较大的导航误差;特别是,艏向的漂移会导致误差的急剧增加。

In addition, considering the user experience, this paper studies a multi-mode smartphone-based navigation algorithm.
此外,考虑到用户体验,本文研究了一种基于智能手机的多模式导航算法。

In order to solve the defects of the traditional PDR algorithm, we propose a novel PDR algorithm, which includes step detection based on state transition, a step length estimation method based on neural network and differential GPS, and multi-mode intelligent recognition, a heading correction method based on zero angular velocity (ZA), and an overall correction method based on lateral velocity limitation (LV), as shown in Figure 2.
为了解决传统的PDR算法的缺陷,我们提出一种新颖的PDR算法,包括基于状态转换步骤检测,步长估算方法基于神经网络和差分GPS和多模智能识别,基于零角速度的航向修正方法(ZA),以及基于横向速度限制(LV)的整体校正方法,如图2。

In the figure, the purple dotted boxes represent the neural networks which are outlined by different features. We estimate the results directly using these neural networks in a PDR algorithm.
在图中,紫色虚线框表示由不同特征勾勒出的神经网络。我们在PDR算法中直接使用这些神经网络来估计结果。

2.1.2总结

首先,介绍传统方法:
首先大致介绍传统PDR的情况,也同时说明了本文的载体(手机)坐标系为右前上,导航坐标系为东北天。当然本文作者描述坐标系的方法是使用传统上都这么做的方法,但是其实还有前右下和北东地。
之后,详细介绍传统方法获得位置航向速度的方法。
最后,简单展示了本文获得这三个信息的方式。

2.2. Step Detection Based On State Transition

第一部分(叙述怎么处理数据的)

During the walking process of the human handheld smartphone, the lifting and undulating of the foot cause the lifting and undulating of the human body. This cyclical fluctuation is reflected in the acceleration of the smartphone, as shown in Figure 3a.
在人类手持智能手机行走过程中,足部的抬升和波动会引起人体的抬升和波动。这种周期性波动反映在智能手机的加速上,如图3a所示

The delt-accNorm in Figure 3a represents the differences between the acceleration modulus and the modulus of the pedestrian’s initial stationary stage, denoted by Equation (5).
图3a中的delt-accNorm表示加速度模量与行人初始静止阶段模量的差值,用式(5)表示。

(这里注意这个东西不是直接使用一个加速度模值,而是使用一个和静止比较的模值)

The purpose of the differences is to reduce the difference of the threshold of step detection between different pedestrians.
差异的目的是减小不同行人的计步器检测阈值的差异。
(他这里的意思是减少每个人在走路时候的差异,但是我个人觉得这里不能减少每个人的差异,例如每个人抬高的高度是不一样的,就算使用这种方式也是不能解决的。但是这样做确实可以消除每个人的在相同持有状态分类下的习惯差异,例如每个人的平端可能多多少少有所不同)

In view of the noise of the accelerometer in the course of motion, the curve burr as shown in Figure 3a will reduce the accuracy of step detection.
由于加速度计在运动过程中存在噪声,如图3a所示的曲线毛刺会降低步进检测的精度。

Thus, we smooth out the interference of acceleration using a moving average filter shown in Equation (6).
因此,我们使用式(6)中所示的移动平均滤波器消除加速度的干扰。

大约就是在一个窗口内取一个均值

The smoothed result is shown in Figure 3b.
平滑后的结果如图3b所示

The valley value f_valley and the peak value f_peak of delt-accNorm are taken as the conditions of step detection.
以delta - accnorm的谷值f_valley和峰值f_peak作为步进检测的条件。
(就是用波峰和波谷来进行检测)

第二部分(具体描述怎么进行状态转化的)

The specific models are as follows:
具体模型如下:
(1) Peak value detection: if fk,peak > THpeak, then the mode-flag is set to 1.
(1)峰值检测:如果fk,峰值>峰值THpeak,则mode-flag设置为1。
(这里就是超过一个阈值就把他当成一个波峰了)

(2) Valley value detection: if fk,valley > THvalley, and the time difference between the last peak and the valley value ∆Tk,peak,valley satisfies ∆Tk,peak,valley > THpeak,valley, then the mode-flag is set to 2.
大约就是说,这里的波谷不仅仅得小于阈值,还得和之前的波峰和波谷有一定的距离

(3) On the basis of (2), the next peak detection is carried out, and if the time difference between the last valley and this peak value (∆Tk,valley,peak) satisfies (∆Tk,valley,peak) > (THvalley,peak), then the mode-flag is set to 3.
我们再次检查到波峰的时候我们也得和之前的波谷进行判断。

(4) When the mode-flag is 3, it represents that we detected one step successfully. Then, the mode-flag is set to 1 and the cycle is repeated.
当mode-flag为3时,表示我们成功检测到一个步骤。然后,将mode-flag设置为1,并重复此循环。

THpeak, THvalley, THpeak,valley, and THvalley,peak are the detection thresholds of peak value, valley value, the time difference between the last peak and the current valley value, and the time difference between the last valley and the current peak value, respectively.
THpeak、THvalley、THpeak、THvalley、THvalley、peak分别为峰值、谷值、最后一个峰值与当前谷值的时间差、最后一个峰值与当前峰值的时间差的检测阈值。

The process of step detection is shown in Figure 4
步进检测的过程如图4所示

第三部分

The difference between the handheld and the foot-fixed smartphone-based navigation algorithm is that the latter has an obvious zero velocity interval [13], while the former has only very short zero velocity moments.
基于手持和基于脚的智能手机导航算法的不同之处在于,后者具有明显的零速度区间[13],而前者只有非常短的零速度时间段。

The zero velocity moment of the handheld smartphone-based navigation is the first sampling period which is less than the modulus value of the initial stationary stage after the peak appears.
基于手持智能手机的导航的零速度时间段是第一个采样周期小于峰值出现后初始稳定阶段的模值。

Meanwhile, the zero velocity flag is set to 1. Moreover, in order to reduce the cumulative errors of velocity, the velocity of that moment is set to zero.
同时,将零速度标志设置为1。为了减小速度的累积误差,将该时刻的速度设为零。

定位相关论文-A Novel Pedestrian Dead Reckoning Algorithm for Multi-Mode Recognition Based on Smartphones相关推荐

  1. PDR (Pedestrian Dead Reckoning)行人航位推算基本原理及实现

    PDR (Pedestrian Dead Reckoning)行人航位推算算法是利用加速度计.磁力计.陀螺仪等多种传感器数据进行处理.对行人行走的步数.步长.方向进行测量和统计,推算出步行者行走轨迹, ...

  2. 论文阅读《Pedestrian Dead Reckoning-Assisted Visual Inertial Odometry Integrity Monitoring》

    目录 摘要 1 介绍 2 背景 3 视觉误差分析和自主完好性监测 3.1 视觉误差分析 3.1.1 不足的特征 3.1.2 光照导致特征跟踪失败 3.1.3 特征分布不均匀 3.1.4 运动特征 3. ...

  3. 【深度学习】步态识别-论文阅读(无参考意义):Cross-View Gait Recognition Based on Feature Fusion

    这里写目录标题 摘要 介绍 相关工作 改进 提出 多尺度特征融合 全局和局部特征融合 特征映射 结论 基于特征融合的跨视图步态识别 摘要 与人脸识别相比,步态识别是最有前途的视频生物特征识别技术之一, ...

  4. Believable Dead Reckoning for Networked Games (为联网游戏而设计的可信的导航预测)

    Believable Dead Reckoning for Networked Games (为联网游戏而设计的可信的导航预测) Curtiss Murphy Alion Scence and Tec ...

  5. Dead Reckoning - 航位推测法

    原文地址: http://www.gamasutra.com/view/feature/131638/dead_reckoning_latency_hiding_for_.php Programmer ...

  6. 阅读软件测试论文:Evaluation of a prioritization algorithm for test suite generation

    文章目录 前言 1 基本信息 2 文章内容 3 几个QA 4 感想 前言 在此阅读软件测试相关论文:Evaluation of a prioritization algorithm for test ...

  7. 基于深度学习实现行人跟踪相关论文总结

    基于深度学习实现行人跟踪相关论文总结 [1] Recurrent YOLO and LSTM-based IR single pedestrian tracking 本文提出了一种基于空间监督的递归卷 ...

  8. KBQA相关论文分类整理:简单KBQA和复杂KBQA

    ©作者 | 蒋锦昊 学校 | 中国人民大学博士生 研究方向 | 知识推理和问答系统 引言 基于知识图谱的问答系统(Knowledge Based Question Answering, KBQA)目标 ...

  9. KBQA相关论文分类整理

    © 作者|蒋锦昊 机构|中国人民大学高瓴人工智能学院直博生 导师|赵鑫教授 研究方向 | 知识推理和问答系统 1.引言 基于知识图谱的问答系统(Knowledge Based Question Ans ...

最新文章

  1. iOS的相对路径和绝对路径
  2. 如何提升网站的信任度,助力网站获取高排名?
  3. 在什么情况下,刘强东会丧失京东的控制权?
  4. ArcGIS 10.2晕渲图+旋转图制作
  5. linux下g++和gcc_Linux中gcc和g ++有什么区别?
  6. linux内存管理_浅谈Linux内存管理
  7. 对于单输入多输出系统matlab,求助!!如何把多输入多输出系统的传函转换为状态空间表达式?...
  8. ANDROID调用webservice带soapheader验证
  9. grep和正则表达式
  10. leetcode之四数相加
  11. 2021-01-29 大数据课程笔记 day9
  12. 解决:启动word等office相关软件时,比较慢,会卡在一步Foxit PDF Creator COM Add-in很久
  13. 不需要PM3/PN532也可以搞定加密卡!利用小米9、小米手环NFC模拟加密门禁卡的傻瓜式保姆级教程
  14. MySQL数据库中主键和候选键的区别?
  15. 地理坐标系:WGS84和BD09互转
  16. 代码进行insmod/rmmod操作
  17. 高德地图自定义定位按钮后搜索周边
  18. libuv之mingw64环境搭建及编译Libuv
  19. 考研视频有点难,以后继续早上锻炼
  20. html图片动态案例,10个强大的纯CSS3动画案例分享

热门文章

  1. 给定数组的子集 Subsets
  2. Mysql 存储引擎和事物基础概念
  3. magent实现memcached集群的一个问题
  4. 关于c中的inline
  5. octet和byte
  6. CSP认证201604-4 游戏[C++题解]:bfs、拆点、迷宫问题加强版、三维数组
  7. Java web对试卷进行单选多选答题进行打分_java + vue 考试系统,适配各种题型,包含微信小程序端...
  8. mysql 存储过程 排序_更改MySQL存储过程“数据库排序规则”的名称
  9. oracle union详解,Oracle中的union和join
  10. NB模块开机、关机、重启的实现