Optimized CUDA Implementation to Improve the Performance of Bundle Adjustment Algorithm on GPUs

The 3D reconstruction pipeline uses the Bundle Adjustment algorithm to refine the camera and point parameters. The Bundle Adjustment algorithm is a compute-intensive algorithm, and many researchers have improved its performance by implementing the algorithm on GPUs. In the previous research work, “Improving Accuracy and Computational Burden of Bundle Adjustment Algorithm using GPUs,” the authors demonstrated first the Bundle Adjustment algorithmic performance improvement by reducing the mean square error using an additional radial distorting parameter and explicitly computed analytical derivatives and reducing the computational burden of the Bundle Adjustment algorithm using GPUs. The naïve implementation of the CUDA code, a speedup of 10× for the largest dataset of 13,678 cameras, 4,455,747 points, and 28,975,571 projections was achieved. In this paper, we present the optimization of the Bundle Adjustment algorithm CUDA code on GPUs to achieve higher speedup. We propose a new data memory layout for the parameters in the Bundle Adjustment algorithm, resulting in contiguous memory access. We demonstrate that it improves the memory throughput on the GPUs, thereby improving the overall performance. We also demonstrate an increase in the computational throughput of the algorithm by optimizing the CUDA kernels to utilize the GPU resources effectively. A comparative performance study of explicitly computing an algorithm parameter versus using the Jacobians instead is presented. In the previous work, the Bundle Adjustment algorithm failed to converge for certain datasets due to several block matrices of the cameras in the augmented normal equation, resulting in rank-deficient matrices. In this work, we identify the cameras that cause rank-deficient matrices and preprocess the datasets to ensure the convergence of the BA algorithm. Our optimized CUDA implementation achieves convergence of the Bundle Adjustment algorithm in around 22 seconds for the largest dataset compared to 654 seconds for the sequential implementation, resulting in a speedup of 30×. Our optimized CUDA implementation presented in this paper has achieved a 3× speedup for the largest dataset compared to the previous naïve CUDA implementation.

Complex Least Squares Adjustment to Improve Tree Height Inversion Problem in PolInSAR

At present,the principal data processing methods involving complex observations are based on two strategies according to characteristics of the observation process,i.e.,step-by-step and direct resolution.However,these strategies have some limitations,e.g.they cannot consider statistical observation error information,redundant observations and so on.This paper applies least squares methods to complex data processing to extend surveying adjustment theory from real to complex number space.We compared the two adjustment criteria for a complex domain in a quantitative way.In order to understand the effectiveness of complex least squares,tree height inversion from PolInSAR data is taken as an example.We firstly established both a complex adjustment function model and a stochastic model for PolInSAR tree height inversion,and then applied the complex least squares method to estimate tree height.Results show that the complex least squares approach is reliable and outperforms other classic tree height retrieval methods;the method is simple and easy to implement.

Algorithm of neural-network about solving nonlinear least squares adjustment by parameters

Study on solving nonlinear least squares adjustment by parameters is one of the most important and new subjects in modern surveying and mapping field . Many researchers have done a lot of work and gained some solving methods. These methods mainly include iterative algorithms and direct algorithms mainly. The former searches some methods of rapid convergence based on which surveying adjustment is a kind of problem of nonlinear programming. Among them the iterative algorithms of the most in common use are the Gauss-Newton method, damped least quares, quasi-Newton method and some mutations etc. Although these methods improved the quantity of the observation results to a certain degree, and increased the accuracy of the adjustment results, what we want is whether the initial values of unknown parameters are close to their real values. Of course, the model of the latter has better degree in linearity, that is to say, they nearly have the meaning of deeper theories researches. This paper puts forward a kind of method of solving the problems of nonlinear least squares adjustment by parameters based on neural network theory, and studies its stability and convergency. The results of calculating of living example indicate the method acts well for solving parameters problems by nonlinear least squares adjustment without giving exact approximation of parameters.

THE LEAST SQUARES COMPREHENSIVE ADJUSTMENT FOR MINE VENTILATION MEASUREMENTS

This paper systematically introduced the method of direct, indirect and free-network adjustments and their application in data process of minewide ventilation measurements. The direct adjustment is suitable for errors collating of the measurements of airflow rates. The indirect method is suitable for the adjustment of ventilation resistance. The free-net method is adapted to the combined adjustment of the measurements of both the ventilation in branches of the air network and the air pressure at nodes of the ventilation network, the partial adjustment is also introduced here to be used for saving the storage locations in computer required for the adjustment for large scale mine ventilation measurements.

A NEW SOLUTION MODEL OF NONLINEAR DYNAMIC LEAST SQUARE ADJUSTMENT

The nonlinear least square adjustment is a head object studied in technology fields. The paper studies on the non derivative solution to the nonlinear dynamic least square adjustment and puts forward a new algorithm model and its solution model. The method has little calculation load and is simple. This opens up a theoretical method to solve the linear dynamic least square adjustment.

Adjustment Model and Colored Noise Compensation of Continuous Observation System

The affection caused by the colored noises should be taken into account to the adjustment model.As useful signals,these colored noises should be accurately identified and extracted by Fourier analysis.A continuous adjustment model is introduced with respect to the colored noises,and then it can be generalized from the finite space to the infinite space so called as Hilbert space.This extension is to provide a new technique to perform the continuous observational system design,Fourier analysis as well as the parameter estimation.It shows that the Gramer’s determinant provides maximization criteria in the system optimization design as well as a rule in diagnosing the adjustment model.Related with the definition of the integral,the least squares solution of the continuous adjustment model becomes the limit of the traditional least squares solution in finite space.Moreover,the influence caused by the colored noises is systematic,but it can be eliminated or compensated by optimally designing the observational system.

Adaptive algorithm for solving different types and different precisions and nonlinear surveying and mapping parameters adjustment

Approximate linear methods and nonlinear methods were adopted usually for solving models of nonlinear surveying and mapping parameters adjustment. But, these iterative algorithms need to compare harsh initial value. A kind of new algorithm-adaptive algorithm based on analyzing the general methods was put forward. The new algorithm has quick rate of convergence and low dependence for initial value, so it can avoid calculating complex second derivative of the target function. The results indicate that its performance is better than those of the others.

Linear-regression models and algorithms based on the Total-Least-Squares principle

In classical regression analysis, the error of independent variable is usually not taken into account in regression analysis. This paper presents two solution methods for the case that both the independent and the dependent variables have errors. These methods are derived from the condition-adjustment and indirect-adjustment models based on the Total-Least-Squares principle. The equivalence of these two methods is also proven in theory.

一种多频多系统周跳探测与修复方法

为满足多频多系统精密单点定位(PPP)数据预处理的实际需求,提出了一种周跳探测与修复方法。针对多频多系统PPP数据处理中各卫星的信号数量不同,利用最小二乘模糊度降相关平差(LAMBDA)原理构建模糊度组合。通过仅使用相位观测值估计的倾斜电子总量(STEC)探测不敏感周跳和漏判周跳。对于周跳修复,将伪距和载波相位的观测值在历元间作差求得周跳的浮点解和协方差阵,然后应用LAMBDA算法搜索周跳整数解并进行修复。通过静态、动态以及磁暴环境下周跳探测和修复实验,验证了所提方法的有效性,多频条件下周跳修复的正确率达到100%。

基于循环神经网络的自适应滤波方法及应用研究

针对目前地震工程研究领域在滤波方法上存在人为因素、峰值突刺、噪声干扰等方面的缺陷,结合递归最小二乘法(RLS)和循环神经网络(RNN)模型,提出了一种自适应滤波的新方法。研究分析表明,该方法通过设置自适应调节滤波器参数以及算法的自我迭代等方式进行滤波,对噪声识别能力和滤波速度上均优于美国地质调查局(United States Geological Survey,USGS)所推荐的传统滤波方法,并可有效降低滤波后对原始波形的失真损坏以及相位提前等问题。同时,运用所提自适应滤波方法将其应用于不同场地类型台站的含速度脉冲近场地震记录,进一步验证了自适应滤波方法的有效性和适用性。研究成果为地震工程领域的滤波分析提出了一种新思路和新方法,也可为地震记录处理及相关应用工作提供参考。

基于变分模态分解和SG滤波的二次谐波降噪

针对可调谐半导体激光器吸收光谱学,在测量低浓度气体过程中二次谐波存在的噪声问题,提出利用变分模态分解(VMD)和Savitzky-Golay滤波(SG滤波)相结合方法(简记为VMD-SG滤波)对含噪信号进行降噪。分析了VMD平衡参数的最优选取,利用所设框长调节因子P,选取SG滤波的最优框长。通过与其他几种降噪算法相比,VMD-SG滤波降噪算法在信噪比、均方根误差、波形恢复、二次谐波特征点误差方面均表现较优。在信噪比(SNR)为-7.8389 dB下该算法SNR提升量为22.047 dB,均方根误差为0.0154,与标准二次谐波相关系数为0.985。结果表明,基于VMD-SG滤波降噪算法的可调谐二极管吸收光谱复合降噪技术,在弱信号方面可以提高信号质量,有利于提高气体反演精度、系统检测灵敏度。

基于Qt的水准网平差系统设计与可视化实现

为满足工程需求提高数据自动处理程度,解决外业水准网测量数据处理问题,本文提出基于Qt的水准网平差系统设计与可视化实现。结合Qt语言的美观简洁和独特的GUI功能优势,依据水准网平差基本原理方法,设计一套界面简单、功能明确合理的平差软件系统,实现基于Qt开发图形水准网平差系统的可视化显示表达。最后,以实际工程为例,验证程序设计界面降低数据输入错误概率、提高运算效率与精度,实验结果验证程序可行性。

基于不同平差软件的GNSS控制测量数据处理方法研究

以某工程场地控制测量项目为研究对象,通过对GNSS控制网布设方法进行优化设计,采用高质量独立基线构成项目GNSS控制网,利用三维无约束平差对观测基线进行粗差检验,保证基线质量。分别采用TBC和COSA 2种GNSS数据处理软件对GNSS控制网进行平差计算,验证了在相同观测数据、相同控制网形、相同起算数据的条件下,不同平差软件精度指标和解算结果会存在一定差异,但成果较差相对较小,故在GNSS控制测量过程中,可采用多种平差软件进行独立计算,成果互为验证,进一步保障控制测量成果的可靠性。

Inversion of vegetation height from Pol In SAR using complex least squares adjustment method

In this paper, we propose the novel method of complex least squares adjustment(CLSA) to invert vegetation height accurately using single-baseline polarimetric synthetic aperture radar interferometry(Pol In SAR) data. CLSA basically estimates both volume-only coherence and ground phase directly without assuming that the ground-to-volume amplitude radio of a particular polarization channel(e.g., HV) is less than ?10 d B, as in the three-stage method. In addition, CLSA can effectively limit errors in interferometric complex coherence, which may translate directly into erroneous ground-phase and volume-only coherence estimations. The proposed CLSA method is validated with Bio SAR2008 P-band E-SAR and L-band SIR-C Pol In SAR data. Its results are then compared with those of the traditional three-stage method and with external data. It implies that the CLSA method is much more robust than the three-stage method.

Nonlinear least square estimation using difference quotient instead of derivative containing different classes of measurements

Using difference quotient instead of derivative, the paper presents the solution method and procedure of the nonlinear least square estimation containing different classes of measurements. In the meantime, the paper shows several practical cases, which indicate the method is very valid and reliable.

Improved Water Network Macroscopic Model Utilising Auto-Control Adjusting Valve by PLS

In order to overcome the low precision and weak applicability problems of the current municipal water network state simulation model, the water network structure is studied. Since the telemetry system has been applied increasingly in the water network, and in order to reflect the network operational condition more accurately, a new water network macroscopic model is developed by taking the auto-control adjusting valve opening state into consideration. Then for highly correlated or collinear independent variables in the model, the partial least squares (PLS) regression method provides a model solution which can distinguish between the system information and the noisy data. Finally, a hypothetical water network is introduced for validating the model. The simulation results show that the relative error is less than 5.2%, indicating that the model is efficient and feasible, and has better generalization performance.

基于SRCKFw-检测的多传感器融合的姿态解算算法

针对惯性导航系统受模型误差和测量异常值误差的影响,姿态解算结果易出现精度差甚至发散的问题,提出了一种基于平方根容积卡尔曼滤波(square-root cubature Kalman filter,SRCKF)w-检测的多传感器姿态融合算法。利用协方差匹配法对SRCKF的新息序列进行自适应调整,经过调整后的新息在迭代过程中会补偿量测噪声方差阵,减小模型误差影响;再利用调整后的新息进行误差探测,提高w-检测的探测精度,并构造观测值替换准则进行误差观测值替换,解决测量异常值误差带来的影响;最后利用SRCKF进行姿态融合,陀螺仪的姿态作为状态方程,经检测替换后的加速度计和磁力计姿态作为量测方程。实验表明,所提算法可以准确估计系统姿态,与传统算法相比解算精度平均可提升62.43%,在不同条件下,算法整体性能均可得到大幅提升,并能快速进行姿态解算,保证解算精度。

基于能量信息耦合梯度调节机制的图像修复算法设计与应用

为了克服当前图像修复算法主要依靠图像的置信度信息来获取优先修复块,忽略了图像的能量信息,导致修复结果中存在不连续及伪吉布斯现象等缺陷,设计了基于能量信息与梯度调节机制的图像修复算法;首先,通过区域能量函数来求取图像的能量信息,以计算待修复块的优先权信息,得到优先修复块;然后,基于图像梯度模值,建立梯度调节机制,以调节样本块的大小,获取与图像纹理相适应的样本块尺寸;引入平方差求和(SSD)函数,以确定最优匹配块;最后,通过像素点间的差异性,构造相似惩罚因子,以更新置信度项,完成图像的修复;实验结果显示,在所提算法中将变化阈值取值为0.5时,较当前图像修复方案而言,所提算法具备更好的修复性能,所得到的修复图像拥有更好的纹理连贯性与更高的结构相似(SSIM)值;当对像素丢失比例达到50%的图像进行修复时,较对照组算法而言,所提算法修复图像的SSIM值分别提高了6.5%和17.5%。

基于变分贝叶斯自适应鲁棒滤波的动对动相对定位算法

针对两个相对运动目标之间基线矢量实时变化导致整周模糊度难以固定,从而导致定位精度偏低的问题,提出一种基于变分贝叶斯自适应鲁棒滤波的北斗四频动对动相对定位算法。结合四频中存在电离层延迟和观测噪声因子小且具有整数特性,建立“宽巷+窄巷+伪距”的无几何模糊度解算模型。设计一种基于变分贝叶斯的自适应鲁棒滤波算法,利用M估计鲁棒化处理观测中的野值,利用变分贝叶斯近似估计时变的观测噪声和系统状态量,以逼近真实后验分布。最后采用最小二乘模糊度降相关平差法进行模糊度固定。实验结果表明,所提算法的模糊度固定率达到99.7%以上,E、N和U三个方向的定位精度均达到厘米级,实现了高精度的动对动相对定位。

BDS-2/BDS-3组合长基线三频RTK算法

针对在长基线动态实时差分定位技术(RTK)中,由于整周模糊度容易受到大气延迟误差的影响,导致其固定率偏低,从而影响定位精度的问题,提出一种基于北斗卫星导航(区域)系统即北斗二号(BDS-2)和北斗三号全球卫星导航系统(BDS-3)组合的长基线定位方法:组合方法中BDS-2采用B1I、B2I和B3I 3种信号,BDS-3采用B1I、B2b和B3I 3种信号,充分利用三频中存在电离层延迟极小且具有整数特性的组合,同时考虑对流层延迟的影响,建立基于弱电离层组合的几何相关中长基线解算模型,并利用卡尔曼(Kalman)模型对天顶对流层延迟误差、三维坐标和弱电离层组合模糊度进行参数估计;再结合部分模糊度固定方法加快模糊度搜索效率,最终实现长基线RTK定位。实验结果表明,该方法能够在长基线下基本实现单历元模糊度固定,定位结果与流动站真值之间的定位偏差可达到厘米级。