仿真假体视觉下基于深度图像的手势识别研究

针对仿真假体视觉下彩色图像和深度图像对于手势识别的不同效果,研究使用Kinect获取彩色图像以及深度图像进行手势识别。通过Kinect提取的骨骼信息与提取的深度图像结合,将人体与背景图像分离,对OpenCV库分离后的图像进行降噪,并进行像素化处理。在不同分辨率(32×32,48×48,64×64)下进行彩色图像和深度图像的手势识别实验。实验结果表明,随着分辨率的增加,手势识别的准确率也不断增加。同一分辨率下,深度图像下的手势识别率总体高于彩色图像下的手势识别率,且在32×32分辨率下,二者差异显著。     

仿真假体视觉下动作视频识别研究

为找寻假体视觉下最优的图像处理策略,设计仿真假体视觉下人类动作识别的心理学物理试验.试验使用三种图像处理策略,包括两种传统的边缘提取算法和一种基于感知检测的视频显著性区域检测算法(Saliency-Aware Geodesic,SAG),分别对UCF-101数据库中30个动作视频进行处理,并将处理后的视频匹配不同仿真光...     

仿真假体视觉下的拼图任务研究

为了研究视觉假体佩戴者完成拼图任务时所需时间与原始图像复杂度、被试性别、光幻视的分辨率以及是否能够识别实验图片之间的关系,借助计算机编程技术,将仿真假体视觉下的拼图任务呈现给视力正常的被试者,并在实验过程中调整实验图片的复杂度(分为简单,中等,复杂三组)和不同的仿真光幻视的分辨率(16×16,24×24,32×32,48×48,64×64,128×128六个分辨率)等参数。实验结果表明,中等难度组的实验用时最多,从简单组到中等难度组识别时间有增长的趋势,从中等难度组到复杂组,识别时间有减少的趋势。男性的平均识别情况好于女性,且随着分辨率的增加,用时整体呈现减少的趋势。     

浮动阈值算法在小目标图像提取中的应用

针对相对复杂图像目标对象的提取问题,介绍了像素图像处理常用算法（背景差法、光流法、帧间差分法）的优点和缺点及适用范围。提出浮动阈值在小目标图像提取的方法及算法流程设计的详细组成。实验表明,这种方法能快速准确提取目标,还原模糊分类后的图像目标,并使背景部分替换成其他颜色,从而实现目标的提取。     

仿真假体视觉下的单目测距研究

为了研究假体佩戴者是否具有异常的立体视觉,影响目标定位,在现实场景中开展仿真假体视觉下单目测距研究。实验采用蒙眼测距方法,被试者佩戴头戴式显示器模拟在仿真假体单眼,仿真假体双眼以及在正常单眼情况观察判断放在地板上不同位置(3.51 m,4.92 m,6.33 m)的目标物体,通过蒙着眼睛轻快地走到他/她所判断的地方,记录并统计分析行走距离。实验结果发现,在三种观察条件下,正常单眼感知精度最高,仿真假体单眼最低。随着物体放置距离的增加,准确度也有细微下降。经统计分析发现,仿真假体单眼与正常单眼对距离判断没显著性差异,说明假体植入者能准确定位地面上的目标。     

用于红外双目视觉的边缘增强算法研究

红外图像的空间关联性强,并且存在一定的非均匀性,因此诸如Canny算子之类的传统边缘提取方法并不适用。本文讨论了像素级和亚像素级结合的边缘检测方法,首先采用脉冲耦合神经网络(pulse coupled neural network,PCNN)方法进行像素级边缘定位,再结合空间灰度矩的方法进行亚像素级边缘细分。该方法能够对高温构件的红外边缘进行快速检测,并能极大提高边缘的定位精度。     

采用神经网络算法的机器视觉成像器

长久以来,智能机器一直是人们梦寐以求的机器,如今,人们的幻想已成为现实。用机器视觉来识别图形不断变化的元件,这需要复杂的算法。神经网络能够模仿类似于人脑细胞的处理功能。由美国通用视觉公司和IBM公司发明的零指令集芯片使得Pulnix公司的ZiCam摄像机能够把图形分解成更小的组元,而神经网络中的分立神经细胞可独立地对这些组元进行建模。换句话说,一个神经细胞可接收有关物体形状的信息,而另一个神经细胞则可记录颜色信息。这些聚合数     

基于Simulink的红外图像处理算法仿真平台

提出一种基于Simulink的、易于工程实现的红外图像处理算法仿真技术.利用Matlab及Simulink这一强大的科学计算和仿真工具,快速地建立针对红外图像处理的仿真平台,并针对教练机的红外视频图像进行了图像滤波及边缘检测算法联合图像处理算法的仿真研究.验证了基于Simulink的红外图像处理算法相对于Visual C++编程语言的实用性、快速性,为今后快速建立针对红外目标图像的实时处理平台打下了基础.     

基于神经网络的双目视觉摄像机标定方法的研究

摄像机标定是精密视觉测量的基础，传统的双目标定位需要建立复杂的数学模型。神经网络可以有效地处理非线性映射问题，本文介绍了一种BP神经网络，可以很好地描述双目视觉中三维空间特征点坐标和2个摄像机对应像点间的非线性关系，并且为了提高网络的学习能力引入了动态因子。将神经网络标定方法与传统的常用标定方法比较，实验结果表明，基于神经网络的双目视觉标定方法能获得较高的标定精度。     

Evolving artificial neural networks

Learning and evolution are two fundamental forms of adaptation. There has been a great interest in combining learning and evolution with artificial neural networks (ANNs) in recent years. This paper: 1) reviews different combinations between ANNs and evolutionary algorithms (EAs), including using EAs to evolve ANN connection weights, architectures, learning rules, and input features; 2) discusses different search operators which have been used in various EAs; and 3) points out possible future research directions. It is shown, through a considerably large literature review, that combinations between ANNs and EAs can lead to significantly better intelligent systems than relying on ANNs or EAs alone     

Digital systems for artificial neural networks

The current art of digital electronic implementation of neural networks is reviewed. Most of this work has taken place as digital simulations on general-purpose serial or parallel digital computers. Specialized neural network emulation systems have also been developed for more efficient learning and use. Dedicated digital VLSI integrated circuits offer the highest near-term future potential for this technology     

Uncertainty in the output of artificial neural networks

Analysis of the performance of artificial neural networks (ANNs) is usually based on aggregate results on a population of cases. In this paper, we analyze ANN output corresponding to the individual case. We show variability in the outputs of multiple ANNs that are trained and "optimized" from a common set of training cases. We predict this variability from a theoretical standpoint on the basis that multiple ANNs can be optimized to achieve similar overall performance on a population of cases, but produce different outputs for the same individual case because the ANNs use different weights. We use simulations to show that the average standard deviation in the ANN output can be two orders of magnitude higher than the standard deviation in the ANN overall performance measured by the Az value. We further show this variability using an example in mammography where the ANNs are used to classify clustered microcalcifications as malignant or benign based on image features extracted from mammograms. This variability in the ANN output is generally not recognized because a trained individual ANN becomes a deterministic model. Recognition of this variability and the deterministic view of the ANN present a fundamental contradiction. The implication of this variability to the classification task warrants additional study.     

Late potential recognition by artificial neural networks

Ventricular late potentials (LPs) are high-frequency low-amplitude signals obtained from signal-averaged electrocardiograms (ECGs) [SAECGs]. LPs are useful in identifying patients prone to ventricular tachycardia (VT), spontaneous or inducible during electrophysiology testing. A combination of self-organizing and supervised artificial neural network (ANN) models was developed to identify patients with a positive electrophysiology (PEP) test for inducible ventricular tachycardia from patients with a negative electrophysiology (NEP) test using LPs. We have added morphology information of vector magnitude waveform to an original set of three time-domain features of LPs, which are total QRS duration (TQRSD), high-frequency low-amplitude signal duration (HFLAD), and root-mean-square voltage (RMSV). Pattern recognition results from an ANN model with this combination feature set are superior to the results from Bayesian classification model based on conventional three time-domain features of SAECG. In order to increase the robustness of the recognition, a filtered QRS offset point is randomly shifted ±8 ms to form a fuzzy training set, which was to simulate the possible error in detecting QRS offset point of filtered SAECG. We also found that nonlinear transformation through the hidden layer of developed ANN model could increase Euclidean distance between PEP and NEP patterns     

Parametric signal restoration using artificial neural networks

The problem of parametric signal restoration given its blurred/nonlinearly distorted version contaminated by additive noise is discussed. It is postulated that feedforward artificial neural networks can be used to find a solution to this problem. The proposed estimator does not require iterative calculations that are normally performed using numerical methods for signal parameter estimation. Thus high speed is the main advantage of this approach. A two-stage neural network-based estimator architecture is considered in which the vector of measurements is projected on the signal subspace and the resulting features form the input to a feedforward neural network. The effect of noise on the estimator performance is analyzed and compared to the least-squares technique. It is shown, for low and moderate noise levels, that the two estimators are similar to each other in terms of their noise performance, provided the neural network approximates the inverse mapping from the measurement space to the parameter space with a negligible error. However, if the neural network is trained on noisy signal observations, the proposed technique is superior to the least-squares estimate (LSE) model fitting. Numerical examples are presented to support the analytical results. Problems for future research are addressed     

Generating ROC curves for artificial neural networks

Receiver operating characteristic (ROC) analysis is an established method of measuring diagnostic performance in medical imaging studies. Traditionally, artificial neural networks (ANN's) have been applied as a classifier to find one “best” detection rate. Recently researchers have begun to report ROC curve results for ANN classifiers. The current standard method of generating ROC curves for an ANN is to vary the output node threshold for classification. Here, the authors propose a different technique for generating ROC curves for a two class ANN classifier. They show that this new technique generates better ROC curves in the sense of having greater area under the ROC curve (AUC), and in the sense of being composed of a better distribution of operating points     

Adaptive convolution kernel for artificial neural networks

Many deep neural networks are built by using stacked convolutional layers of fixed and single size (often 3 × 3) kernels. This paper describes a method for learning the size of convolutional kernels to provide varying size kernels in a single layer. The method utilizes a differentiable, and therefore backpropagation-trainable Gaussian envelope which can grow or shrink in a base grid. Our experiments compared the proposed adaptive layers to ordinary convolution layers in a simple two-layer network, a deeper residual network, and a U-Net architecture. The results in the popular image classification datasets such as MNIST, MNIST-CLUTTERED, CIFAR-10, Fashion, and “Faces in the Wild” showed that the adaptive kernels can provide statistically significant improvements on ordinary convolution kernels. A segmentation experiment in the Oxford-Pets dataset demonstrated that replacing ordinary convolution layers in a U-shaped network with 7 × 7 adaptive layers can improve its learning performance and ability to generalize.     

Feedforward associative memory switched-capacitor artificial neural networks

Observed spatiotemporal firings of biological neurons have lead many researchers to believe that the rate of firings of these biological neurons is what conveys neuronal information in the brain. In this paper we seek to highlight parallels between biological neurons and observed effects in real neurons, with artificial neurons implemented as switched-capacitor structures. One such effect is the heavy use of lateral inhibition observed in the brain that is often modeled by winner-take-all analog circuits. This paper introduces a novel winner-take-all circuit using switched capacitors that truly mimics this effect seen in biological systems. In addition, we show how switched-capacitor structures can also cater to both binary and bipolar coding of input data vectors, as required by many artificial neural network paradigms today. Applications of switched-capacitors artificial neural networks to pattern recognition and character recognition problems using feedforward associative neural networks are also discussed, and two examples are provided. Simulations using both HSPICE and SWITCAP2 confirm all our expectations.     

Edge crossing minimisation using artificial neural networks

An artificial neural network is proposed for solving the NP-complete problem of edge crossing minimisation optimally and in parallel     

Synthesis of artificial neural networks by transconductors only

Hardware implementation of artificial neural networks has been attracting great attention recently. In this work, the analog VLSI implementation of artificial neural networks by using only transconductors is presented. The signal flow graph approach is used in synthesis. The neural flow graph is defined. Synthesis of various neural network configurations by means of neural flow graph is described. The approach presented in this work is technology independent. This approach can be applied to new neural network topologies to be proposed or used with transconductors designed in future technologies.