Energy-Based Geometric Multi-model Fitting

Geometric model fitting is a typical chicken-&-egg problem: data points should be clustered based on geometric proximity to models whose unknown parameters must be estimated at the same time. Most existing methods, including generalizations of RANSAC, greedily search for models with most inliers (within a threshold) ignoring overall classification of points. We formulate geometric multi-model fitting as an optimal labeling problem with a global energy function balancing geometric errors and regularity of inlier clusters. Regularization based on spatial coherence (on some near-neighbor graph) and/or label costs is NP hard. Standard combinatorial algorithms with guaranteed approximation bounds (e.g. α-expansion) can minimize such regularization energies over a finite set of labels, but they are not directly applicable to a continuum of labels, e.g. R²{\mathcal{R}}^{2} in line fitting. Our proposed approach (PEaRL) combines model sampling from data points as in RANSAC with iterative re-estimation of inliers and models’ parameters based on a global regularization functional. This technique efficiently explores the continuum of labels in the context of energy minimization. In practice, PEaRL converges to a good quality local minimum of the energy automatically selecting a small number of models that best explain the whole data set. Our tests demonstrate that our energy-based approach significantly improves the current state of the art in geometric model fitting currently dominated by various greedy generalizations of RANSAC.     

A personalized e-course composition based on a genetic algorithm with forcing legality in an adaptive learning system

This paper proposes a personalized e-course composition based on a genetic algorithm with forcing legality (called GA^?) in adaptive learning systems, which efficiently and accurately finds appropriate e-learning materials in the database for individual learners. The forcing legality operation not only reduces the search space size and increases search efficiency but also is more explicit in finding the best e-course composition in a legal solution space. In serial experiments, the forcing legality operation is applied in Chu et al.'s the particle swarm optimization (called PSO^?) and Dheeban et al.'s the improved particle swarm optimization (called RPSO^?) to show the forcing legality can speed up the computational time and reduce the computational complexity of algorithm. Furthermore, GA^? regardless of the number of students or the number of materials in the database, to compose a personalized e-course within a limited time is much more efficient and accurate than PSO^? and RPSO^?. For the experiment increasing the number of students to 1200, the average improvement ratios of errors (learning concept error, materials difficulty error, learning time error), fitness value, stability, and execution time are above 96%, 79%, 90%, and 10%, respectively. For the experiment increasing the number of materials to 500 and the execution time set to the shortest execution time of RPSO^?, the average improvement ratios of errors (learning concept error, materials difficulty error, learning time error), fitness value, and stability are above 97%, 51%, and 80%, respectively. Therefore, GA^? is able to enhance the quality of personalized e-course compositions in adaptive learning environments.     

Curve-fit analysis of chlorophyll fluorescence spectra: Application to nutrient stress detection in sunflower

The in vivo laser-induced chlorophyll fluorescence (LICF) spectra of healthy and nutrient-deficient sunflower plants were measured on a Jobin Yvon monochromator with He---Ne laser excitation. To correctly determine the peak center and to evaluate the relative contributions of the bands in the total fluorescence spectrum, the steady state LICF spectra were analyzed with a nonlinear iterative procedure using Gaussian, Lorentzian, Pearson, Voigt, and exponential Gaussian spectral functions. It was observed that curve fitting performed by using two Gaussian peaks centered at 690 and 730 nm usually fits well to the chlorophyll fluorescence spectra. After curve fitting, the mean peak centers of the red and far-red chlorophyll bands of control plants were observed at 688.2 and 725.4 nm, respectively. A blue shift of as much as 9 nm in the peak position of the far-red band was observed with nutrient stress, whereas the shift in position of the red band was only of the order of a few nanometers. Further, the width at half maximum of the far-red band was found to increase by as much as 20 nm with nutrient stress. Curve fitting could thus separate out the red and far-red fluorescence spectra from a pair of normally distributed curves centered at 690 and 730 nm wavelengths, thereby differentiating the effects due to reabsorption from those due directly to changes in photosynthetic electron transport. The F690/F730 fluorescence intensity ratio obtained from curve-fitted parameters was found to be more sensitive to plant stress than were fluorescence values alone. Results indicate that a curve-fitting analysis of LICF spectra using Gaussian spectral functions is a very useful and sensitive method of evaluating spectral features from a statistical point of view and for accurate determination of contributions from constituent bands in the whole leaf fluorescence spectrum.     

基于高斯拟合的自适应多阈值分割算法

提出一种新的基于邻域平均直方图的快速高斯拟合函数参数估计的算法，对形态滤波后的邻域平均直方图进行峰谷检出，对各峰区数据单独进行截断数据统计分析确定拟合参数，较好地避免了多峰交叠所带来的干扰，从而实现了复杂图像的自适应多阈值分割。实验表明，该算法能有效提高拟合精度和分割的鲁棒性。     

A comparison of isokinetic lifting strength with static strength and maximum acceptable weight with special reference to speed of lifting

A laboratory study was conducted to determine the effects of the speed of lifting and box size on isokinetic strength and to compare isokinetic lifting strengths with static lifting strengths and psychophysically determined maximum acceptable weights. Nine male college students lifted three different boxes (250, 380 and 510 mm wide) from the floor to a bench height of 0.8 m using a free-style lifting technique at a rate of 0.2 lifts min^?1. For each lifting task static strength was measured at the origin of lift. Isokinetic lifting strength was measured at 0.41,0.51 and 0.6 ms^?1 using a Biokinetic ergometer and attaching boxes to the load cell. Ratings of perceived exertion were recorded for the low back. There was a progressive decrease in mean and peak isokinetic lifting strengths both with an increase in lifting speed and with an increase in box width (p<0.01). The lifting speed had a much greater effect (29% and 27%) than the box width (18% and 15%) on mean and peak isokinetic lifting strengths. However, high speed lifting was perceived subjectively to be less stressful (RPE = 10.7) than slow speed lifting (RPE = 12.7). Static strength and maximum acceptable weight had higher correlations with mean isokinetic strengm (r = 0.65 and 0.82) than with peak isokinetic strength (r = 0.52 and 0.73). At 0.41 ms^?1, mean isokinetic strength was 6% greater than the mean static strength (p ≥0,05). Extrapolation of mean isokinetic strength data showed that at 0.73 ms^?1 the estimated mean isokinetic strengths were within 6% of maximum acceptable weights. It is concluded that isokinetic strength is highly dependent upon die speed of lifting. At a slow speed (0.41 ms^?1), mean isokinetic strength is equal to mean static strength; and, at a high speed (0.73 ms^?1), it appears to be equal to the maximum acceptable weight. It is recommended that both speed of lifting and box width should be controlled carefully to simulate job-specific isokinetic lifting strength.     

基于脚部惯性传感数据的人员运动速度识别方法

现有的基于脚部惯性传感数据的人员运动速度估计方法只能对人员低速行走时的速度进行有效的估计。为了采用脚步惯性传感数据识别人员快速行走以及跑步时的速度,该文提出了一种利用单步统计特征进行速度识别的方法。该方法利用脚部惯性传感器对人员在不同速度下运动的惯性数据进行采集,采用峰值检测的方法对数据进行单步划分,最后从单步数据中提取65维统计特征分别采用最小二乘法（LS）、支持向量机（SVM）、K近邻（KNN）、线型贝叶斯正态分类器（LDC）4种常见的机器学习分类方法对人员运动速度进行识别。经实验验证,所建议的方法中采用SVM分类器的识别率高达96.3%,所以采用该方法可以有效的识别人员的运动速度。     

Learning GP-BayesFilters via Gaussian process latent variable models

GP-BayesFilters are a general framework for integrating Gaussian process prediction and observation models into Bayesian filtering techniques, including particle filters and extended and unscented Kalman filters. GP-BayesFilters have been shown to be extremely well suited for systems for which accurate parametric models are difficult to obtain. GP-BayesFilters learn non-parametric models from training data containing sequences of control inputs, observations, and ground truth states. The need for ground truth states limits the applicability of GP-BayesFilters to systems for which the ground truth can be estimated without significant overhead. In this paper we introduce GPBF-Learn, a framework for training GP-BayesFilters without ground truth states. Our approach extends Gaussian Process Latent Variable Models to the setting of dynamical robotics systems. We show how weak labels for the ground truth states can be incorporated into the GPBF-Learn framework. The approach is evaluated using a difficult tracking task, namely tracking a slotcar based on inertial measurement unit (IMU) observations only. We also show some special features enabled by this framework, including time alignment, and control replay for both the slotcar, and a robotic arm.     

A sampling approach for skyline query cardinality estimation

A skyline query returns a set of candidate records that satisfy several preferences. It is an operation commonly performed to aid decision making. Since executing a skyline query is expensive and a query plan may combine skyline queries with other data operations such as join, it is important that the query optimizer can quickly yield an accurate cardinality estimate for a skyline query. Log Sampling (LS) and Kernel-Based (?KB) skyline cardinality estimation are the two state-of-the-art skyline cardinality estimation methods. LS is based on a hypothetical model A(log(n)) ^B . Since this model is originally derived under strong assumptions like data independence between dimensions, it does not apply well to an arbitrary data set. Consequently, LS can yield large estimation errors. KB relies on the integration of the estimated probability density function (PDF) to derive the scale factor ?? _ds . As the estimation of PDF and the ensuing integration both involve complex mathematical calculations, KB is time consuming. In view of these problems, we propose an innovative purely sampling-based (PS) method for skyline cardinality estimation. PS is non-parametric. It does not assume any particular data distribution and is, thus, more robust than LS. PS does not require complex mathematical calculations. Therefore, it is much simpler to implement and much faster to yield the estimates than KB. Extensive empirical studies show that for a variety of real and synthetic data sets, PS outperforms LS in terms of estimation speed, estimation accuracy, and estimation variability under the same space budget. PS outperforms KB in terms of estimation speed and estimation variability under the same performance mark.     

Sub-pixel precision image matching for measuring surface displacements on mass movements using normalized cross-correlation

This study evaluates the performance of two fundamentally different approaches to achieve sub-pixel precision of normalised cross-correlation when measuring surface displacements on mass movements from repeat optical images. In the first approach, image intensities are interpolated to a desired sub-pixel resolution using a bi-cubic interpolation scheme prior to the actual displacement matching. In the second approach, the image pairs are correlated at the original image resolution and the peaks of the correlation coefficient surface are then located at the desired sub-pixel resolution using three techniques, namely bi-cubic interpolation, parabola fitting and Gaussian fitting. Both principal approaches are applied to three typical mass movement types: rockglacier creep, glacier flow and land sliding. In addition, the influence of pixel resolution on the accuracies of displacement measurement using image matching is evaluated using repeat images resampled to different spatial resolutions. Our results show that bi-cubic interpolation of image intensity performs best followed by bi-cubic interpolation of the correlation surface. Both Gaussian and parabolic peak locating turn out less accurate. By increasing the spatial resolution (i.e. reducing the ground pixel size) of the matched images by 2 to 16 times using intensity interpolation, 40% to 80% reduction in mean error in reference to the same resolution original image could be achieved. The study also quantifies how the mean error, the random error, the proportion of mismatches and the proportion of undetected movements increase with increasing pixel size (i.e. decreasing spatial resolution) for all of the three mass movement examples investigated.     

Linear learning control of robot motion

For the trajectory following problem of a robot manipulator, a new linear learning control law, consisting of the conventional proportional-integral-differential (PID) control law, with respect to position tracking error, and an iterative learning term is provided. The learning part is a linear feedback control of position, velocity, and acceleration errors (PDD²). It has been shown that, under the proposed learning control, the position, velocity, and acceleration tracking errors are asymptotically stable in the presence of highly nonlinear dynamics. The proposed control is robust in the sense that exact knowledge about nonlinear dynamics is not required except for the bounding functions on their magnitudes. Further, neither is linear approximation of nonlinear dynamics nor repeatability of robot motion required.