基于矩形特征和改进Adaboost的手势检测

为了实时、精确地从视频流中检测和识别出特定手势,提出一种矩形特征描述手势,给出快速计算方法和手势类可分离性的评价方法。为了避免分类器的过度训练问题,提出一种基于此方法的改进的Adaboost算法。实验结果表明,矩形特征能够产生可靠的检测器,对手势的姿态变化较敏感。在摄像头实时捕获视频中,其检测手势实时性较好,对复杂背景和噪声有较强的适应能力,当手势旋转角度小时,正确检测率可以达到95%以上。     

基于手势识别的机器人人机交互技术研究

研究了基于视觉的动态手势识别技术,采用基于肤色的高斯模型与改进的光流场跟踪算法结合的方 法,实现了复杂背景下实时的手势跟踪,具有快速和准确的特点,且具有较好的鲁棒性．对于动态手势识别器,采 用了隐马尔可夫模型（HMM）作为训练识别算法．考虑到动态手势特征本身的一些特点,对HMM 参数优化算法重 估式加以修正,调整了算法比例因子,从而推导了最佳状态链的确定算法、HMM 参数优化算法．最后将研究开发 的动态手势识别算法成功地应用到了基于网络的远程机器人控制系统中．     

电容式油量测量系统的姿态误差分析

针对目前飞机上电容式油量传感器测量燃油时产生的姿态误差问题进行了研究,以飞机上的规则油箱为例,从理论上详细地分析了电容式油量测量系统的姿态误差,并绘制出其误差曲线,给出了减小姿态误差的方法,最后提出了在姿态误差最小的准则下确定传感器的最佳安装位置;文中提出的减小姿态误差的方法,不仅有利于提高燃油测量系统的可靠性和安全性,而且可以改善燃油测量系统的维修性,使燃油量测量技术跃上了一个新的台阶。     

基于肤色检测技术的手势分割

手语是聋哑人使用的语言。在手势输入设备发展二十多年以后,许多人仍然发现在与聋哑人的互动交流中存在着一定的困难。因此应该努力使计算机适应自然的交流方式。文中的目的是利用肤色检测程序从整个图像上获取有意义的手势区域,并处理不同室内背景和照明情况下的手势,为进一步的手语识别做准备。对几个静态手势的检测过程进行了实验评估,并就结果进行了讨论。     

融合多尺度特征的复杂手势姿态估计网络

目的 基于单幅RGB图像的手势姿态估计受手势复杂性、手指特征局部自相似性及遮挡问题的影响,导致手势姿态估计准确率低。为此,提出一种面向单目视觉手势姿态估计的多尺度特征融合网络。方法 1）采用ResNet50(50-layer residual network)模块从RGB图像提取不同分辨率特征图,通过通道变换模块显式地学习特征通道间的依赖关系,增强重要的特征通道信息,弱化次要的特征通道信息。2）在全局回归模块中,通过设计节点间的连接方式融合不同分辨率特征图,以便充分利用图像的细节与整体信息。采用局部优化模块继续提取更深层的特征信息,获得手部关节点的高斯热图,以此修正遮挡等原因造成部分关节点回归不准确的问题。3）计算经通道变换模块处理后的最小特征图,通过全局池化和多层感知机处理该特征图以获得手势类别和右手相对于左手的深度。4）综合以上结果获得最终的手势姿态。结果 采用InterHand2.6M和RHD(rendered handpose dataset)数据集训练多尺度特征融合网络,评估指标中根节点的平均误差和关节点的平均误差,均低于同类方法,且在一些复杂和遮挡的场景下鲁棒性更高。在In...     

A two stream convolutional neural network with bi-directional GRU model to classify dynamic hand gesture

Dynamic hand gesture recognition is still an interesting topic for the computer vision community. A set of feature vectors can represent any hand gesture. A Recurrent Neural Network (RNN) can recognize these feature vectors as a hand gesture that analyzes the temporal and contextual information of the gesture sequence. Thus, we proposed a hybrid deep learning framework to recognize dynamic hand gestures. In the Hybrid model GoogleNet is pipelined with a Bidirectional GRU unit to recognize the dynamic hand gesture. Dynamic hand gestures consist of many frames, and features of each frame need to be extracted to get the temporal and dynamic information of the performed gesture. As RNN takes input as a sequence of feature vectors, we extract features from videos using pretrained GoogleNet. As Gated Recurrent Unit is one of the variants of RNN to classify the sequential data, we created a feature vector that corresponds to each video and passed it to the bidirectional GRU (BGRU) network to classify the gestures. We evaluate our model on four publicly available hand gesture datasets. The proposed method performs well and is comparable with the existing methods. For instance, we achieved 98.6% accuracy on Northwestern University Hand Gesture(NWUHG), 99.6% on SKIG, 99.4% on Cambridge Hand Gesture (CHG) datasets respectively. We performed our experiments on DHG14/28 dataset and achieved an accuracy of 97.8% with 14-gesture classes and 92.1% on 28-gesture classes. DHG14/28 dataset contains skeleton and depth data, and our proposed model used depth data and achieved comparable accuracy.     

Human‐Finger Electronics Based on Opposing Humidity‐Resistance Responses in Carbon Nanofilms

Carbon nanomaterials have excellent humidity sensing properties. Here, it is demonstrated that multiwalled carbon‐nanotube (MWCNT)‐ and reduced‐graphene‐oxide (rGO)‐based conductive films have opposite humidity/electrical resistance responses: MWCNTs increase their electrical resistance (positive response) and rGOs decrease their electrical resistance (negative response). The authors propose a new phenomenology that describes a “net”‐like model for MWCNT films and a “scale”‐like model for rGO films to explain these behaviors based on contributions from junction resistances (at interparticle junctions) and intrinsic resistances (of the particles). This phenomenology is accordingly validated via a series of experiments, which complement more classical models based on proton conductivity. To explore the practical applications of the converse humidity/resistance responses, a humidity‐insensitive MWCNT/rGO hybrid conductive films is developed, which has the potential to greatly improve the stability of carbon‐based electrical device to humidity. The authors further investigate the application of such films to human‐finger electronics by fabricating transparent flexible devices consisting of a polyethylene terephthalate substrate equipped with an MWCNT/rGO pattern for gesture recognition, and MWCNT/rGO/MWCNT or rGO/MWCNT/rGO patterns for 3D noncontact sensing, which will be complementary to existing 3D touch technology.     

Microsphere‐Assisted Robust Epidermal Strain Gauge for Static and Dynamic Gesture Recognition

A novel and robust epidermal strain gauge by using 3D microsphere arrays to immobilize, connect, and protect a multiwalled carbon nanotubes (MWNTs) pathway is presented. During the solvent deposition process, MWNTs sedimentate, self‐assemble, and wrap onto surface of polystyrene (PS) microspheres to construct conductive networks, which further obtain excellent stretchability of 100% by combining with commercially used elastomer. Benefiting from its 3D conductive pathway defined by microspheres, immobilized MWNT (I‐MWNT) network can be directly used in practical occasions without further packaging and is proved by tape tests to be capable of defend mechanical damage effectively from external environment. By parameter optimization, the strain sensor with 3 µm PS spheres obtains stable resistive responses for more than 1000 times, and maintains its gauge factor (GF) of 1.35. This thin‐film conductive membrane built by this effective construction method can be easily attached onto fingers of both robot and human, and is demonstrated in sensitive epidermal strain sensing and recognizing different hand gestures effectively, in static and dynamic modes, respectively.     

非规则油箱油量的测量方法

针对采用电容传感器测量油箱中油量的传统测量方法的弊端,提出了非规则油箱的数学模型建立方法及其油量的计算方法,从理论上详细地分析了利用线性电容传感器测量非规则油箱时产生姿态误差的原理,给出了姿态误差的修正方法,最后提出了在姿态误差最小的准则下确定传感器的最佳安装位置;该研究对提高非规则油箱的油量测量精度具有很大的实用价值.     

基于MEMS组合模块的姿态检测系统设计

在研究两轮自平衡电动车的平衡控制的基础上,设计了基于MEMS组合模块的姿态测量系统.通过对各MEMS传感器输出信号特征的提取、分析,将离散化低通滤波器模型与互补滤波的思想相结合,巧妙地构造出一种便于本系统实现的数据融合模型.同时,提供了一种基于MEMS组合模块姿态检测的低成本、可行性方案.