基于单目视觉的FMS人体姿态识别

针对部队日常体能训练中引入功能性动作筛查(FMS)的困难,提出一种基于单目视觉的自动化解决方案.通过改进现有人体姿态识别算法,再结合余弦相似性完成FMS动作提取.分析现有算法的弊端,提出利用帧间光流提高人体姿态识别精度的解决思路,并验证其可行性.     

基于多相机的人脸姿态识别

主动形状模型(ASM)算法被用来进行人脸特征点的精确定位,然后在多相机测量的图像中进行特征点的立体匹配,利用双目视觉和相机三维测距技术可以确定人脸特征点的空间三维位置,从而利用这些特征点的相对位置确定出人脸的姿态。实验结果显示,用该方法进行人脸姿态识别能取得比二维识别更高的精确度。     

基于区域形状与运动特征的实时行为识别

提出了一种基于推广的Hu不变矩特征的实时行为识别方法。首先,对Hu不变矩进行改进,使其在离散情况下同时具有平移、旋转和比例不变性。然后,结合运动目标的速度将目标行为刻画成结合Hu矩新特征和速度特征的13维特性向量。其中,Hu矩新特征表征了行为的区域形状特性,速度特征反映了行为的运动特性。随后采用预先定义的一些行为作为先验知识样本训练支持向量机,并最后使用其对待检测行为进行分类以达到行为识别的效果。所提方法计算效率高,能够实时检测人体行为。在处理实拍视频数据的实验中,该方法表现出了理想的处理效率以及识别精度。     

基于深度信念网络的湖羊维持行为识别

为提升湖羊的福利化养殖水平和推动动物福利事业健康发展,提出了基于深度信念网络(Deep Belief Network, DBN) 的湖羊维持行为识别方法。挑选6只湖羊佩戴装有姿态传感器的颈环,经数据采集和整理,构建了包括58680个样本的湖羊维持行为数据集,记录了湖羊卧息、采食、饮水、反刍4种维持行为,结合错误率和重构误差两项评价指标,构建了逐层贪婪二次划分算法的DBN识别模型,经训练后,在测试集上与传统的BP神经网络(BPNN)、随机森林(RF)、支持向量机(SVM)模型 进行对比分析?同时对湖羊进行分组识别对比分析,结果表明:本文方法明显优于其他三种方法,4种维持行为的平均识别精度和灵敏度分别为0.9916和0.9915,验证了该方法在湖羊维持行为识别上的有效性。本研究结果可为湖羊的福利化养殖、 行为学研究、异常行为识别及疾病预警提供技术支持     

基于高分辨率网络的人体姿态估计方法

人体姿态估计在人机交互和行为识别应用中起着至关重要的作用,但人体姿态估计方法在特征图尺度变化中难以预测正确的人体姿态.为了提高姿态估计的准确性,将并行网络多尺度融合方法和生成高质量特征图的方法结合进行人体姿态估计(RefinedHRNet).在人体检测基础之上,采用并行网络多尺度融合方法在阶段内采用空洞卷积模块来扩大感...     

基于CNN的人体姿态识别

姿态识别是人机交互中重要的研究课题之一，随着机器学习与神经网络的发展，研究的方式和成果趋于多样化，姿态识别的应用价值也日趋广泛。本文通过构建卷积神经网络模型，该模型共有11层，在对采样的数据集中5种人体姿态进行卷积与池化操作，最后进入全连接层进行分类，从而完成对数据集的训练和识别。结果显示，相较于机器学习方法，该模型的识别性能更加优秀，且免去了复杂的特征提取方式设计，让网络自身提取特征进行识别分类，效果更好。     

基于人体图像生成的姿态无关人物识别

人物识别技术能够使机器人具备对用户身份识别的能力,从而有效提高机器人的智能交互水平。人物识别面临的主要挑战之一是姿态的变化对人物身份特征提取的影响。针对该问题,提出基于人体图像生成的姿态无关人物识别方法,通过生成与库中目标人物相同姿态的人体图像,消除姿态变化对人物外观特征造成的影响。该方法首先利用人体分割图将人体区域与背景分离,尽量降低复杂多变的背景对人物外观特征的干扰;然后在目标姿态的引导下生成与目标图像姿态一致的人物图像;最后设计了一个特征融合模块将源图像和生成图像的身份特征进行融合,提取姿态无关的鲁棒身份特征用于人物识别。此外,为更好地区分不同的人物,在训练中生成相同姿态的负样本,对约束模型学习更为细粒的可鉴别性身份特征。人物识别和人体图像生成的实验结果验证了该方法的有效性。     

姿态情绪识别研究综述

从姿态信息采集、姿态情绪特征提取、姿态情绪识别算法和姿态情绪数据库几个方面对国内外姿态情绪识别研究进行了全面的总结,分析了姿态情绪识别研究存在的难点和挑战,提出姿态情绪识别的关键是姿态情绪特征提取和姿态情绪数据库的建立,最后探讨了姿态情绪识别研究的发展方向.     

基于特征向量匹配的舞蹈实时姿态分析方法

针对人体运动姿态可自由编辑的特性,提取人体骨架模型,建立人体运动姿态模型库,分析了运动捕捉系统在舞蹈训练中的应用,提出基于特征平面间相似性匹配的方法来计算模型各部件间运动数据参数。经过验证,该方法对于人体姿态的分析具有较高的准确性、鲁棒性,使得舞蹈演员能够准确的对比出与标准舞蹈动作的差异,为舞蹈者进行科学舞蹈训练提供了理论支持。     

基于多视图半监督学习的人体行为识别

由于人的行为在本质上的复杂性,单一行为特征视图缺乏全面分析人类行为的能力.文中提出基于多视图半监督学习的人体行为识别方法.首先,提出3种不同模态视图数据,用于表征人体动作,即基于RGB模态数据的傅立叶描述子特征视图、基于深度模态数据的时空兴趣点特征视图和基于关节模态数据的关节点投影分布特征视图.然后,使用多视图半监督学习框架建模,充分利用不同视图提供的互补信息,确保基于少量标记和大量未标记数据半监督学习取得更好的分类精度.最后,利用分类器级融合技术并结合3种视图的预测能力,同时有效解决未标记样本置信度评估问题.在公开的人体行为识别数据集上实验表明,采用多个动作特征视图融合的特征表示方法的判别力优于单个动作特征视图,取得有效的人体行为识别性能.     

基于选择性集成旋转森林的人体行为识别算法*

人体行为识别中的一个关键问题是如何表示高维的人体动作和构建精确稳定的人体分类模型.文中提出有效的基于混合特征的人体行为识别算法.该算法融合基于外观结构的人体重要关节点极坐标特征和基于光流的运动特征,可更有效获取视频序列中的运动信息,提高识别即时性.同时提出基于帧的选择性集成旋转森林分类模型(SERF),有效地将选择性集成策略融入到旋转森林基分类器的选择中,从而增加基分类器之间的差异性.实验表明SERF模型具有较高的分类精度和较强的鲁棒性.     

基于注意力机制的时间分组深度网络行为识别算法

受人脑视觉感知机制启发,在深度学习框架下提出基于注意力机制的时间分组深度网络行为识别算法.针对局部时序信息在描述持续时间较长的复杂动作上的不足,使用视频分组稀疏抽样策略,以更低的成本进行视频级时间建模.在识别阶段引入通道注意力映射,进一步利用全局特征信息和捕捉分类兴趣点,执行通道特征重新校准,提高网络的表达能力.实验表明,文中算法在UCF101、HMDB51数据集上的识别准确率较高.     

Graph-based feature recognition for injection moulding based on a mid-surface approach

This paper presents a novel CAD feature recognition approach for thin-walled injection moulded and cast parts in which moulding features are recognised from a mid-surface abstraction of the part geometry. The motivation for the research has been to develop techniques to help designers of moulded parts to incorporate manufacturing considerations into their designs early in the design process. The main contribution of the research has been the development of an attributed mid-surface adjacency graph to represent the mid-surface topology and geometry, and a feature recognition methodology for moulding features. The conclusion of the research is that the mid-surface representation provides a better basis for feature recognition for moulded parts than a B-REP solid model. A demonstrator that is able to identify ribs, buttresses, bosses, holes and wall junctions has been developed using C++, with data exchange to the CAD system implemented using ISO 10303 STEP. The demonstrator uses a commercial algorithm (I-DEAS) to create the mid-surface representation, but the feature recognition approach is generic and could be applied to any mid-surface abstraction. The software has been tested on a range of simple moulded parts and found to give good results.     

Prediction of human reach posture using a neural network for ergonomic man models

For proper evaluation of operator's usability through ergonomic man models, accurate prediction of human reach is one of the essential functions that those models should possess. This study examined the applicability of artificial neural networks to the prediction of human reach posture. The three-dimensional motion trajectories of the joints of upper limb (shoulder, elbow, and wrist) in the right arm from 5 percentile female to 95 percentile male were obtained through a motion analysis system that photographed actual human reach. The data obtained were divided into two data sets — training data set and test data set. The backpropagation method being usually used for a pattern associator was employed as a tool for predicting such human movements. Comparisons between prediction and real measurements were made using a pairwise t-test, and no significant differences were found between the two data sets for all the joints considered. Thus, the neural network approach adopted in this study showed a very promising prediction capability of human reach and it is, therefore, expected that this method be used to accurately simulate human reach better than existing heuristic or analytic methods as well as to improve a human modelling capability in general.     

Applying incremental Deep Neural Networks-based posture recognition model for ergonomics risk assessment in construction

Monitoring and assessing awkward postures is a proactive approach for Musculoskeletal Disorders (MSDs) prevention in construction. Machine Learning models have shown promising results when used in recognition of workers’ posture from Wearable Sensors. However, there is a need to further investigate: i) how to enable Incremental Learning, where trained recognition models continuously learn new postures from incoming subjects while controlling the forgetting of learned postures; ii) the validity of ergonomics risk assessment with recognized postures. The research discussed in this paper seeks to address this need through an adaptive posture recognition model– the incremental Convolutional Long Short-Term Memory (CLN) model. The paper discusses the methodology used to develop and validate this model’s use as an effective Incremental Learning strategy. The evaluation was based on real construction workers’ natural postures during their daily tasks. The CLN model with “shallow” (up to two) convolutional layers achieved high recognition performance (Macro F1 Score) under personalized (0.87) and generalized (0.84) modeling. Generalized CLN model, with one convolutional layer, using the “Many-to-One” Incremental Learning scheme can potentially balance the performance of adaptation and controlling forgetting. Applying the ergonomics rules on recognized and ground truth postures yielded comparable risk assessment results. These findings support that the proposed incremental Deep Neural Networks model has a high potential for adaptive posture recognition. They can be deployed alongside ergonomics rules for effective MSDs risk assessment.     

基于不变矩特征和神经网络的步态识别

步态识别是利用人体步行的方式来识别人的身份.近年来,步态作为一种生物特征识别技术已引起越来越多人们的兴趣.本文提出了一种简单有效的步态识别算法,首先通过背景差方法得到运动人体轮廓,然后利用不变矩描述轮廓特征,最后用BP神经网络方法来进行模板匹配,实现人的身份识别.     

基于加速度特征的可拓动作识别方法

针对基于图像和视频的动作识别系统具有特征采集设备复杂、视角固定和需要采集多视角图像等缺点,提出基于加速度特征的可拓动作识别方法。该方法利用物体向不同方向运动时,其关键部位点的三轴加速度具有一定区分度的特点,结合可拓识别方法,实现动作识别。在构建的手臂动作识别系统中,测得动作识别率可达94.4%。该方法可应用于智能监控、医疗电子等领域。     

Sleep posture recognition based on machine learning: A systematic review

Background:In recent years, the application of artificial intelligence in the field of sleep medicine has rapidly emerged. One of the main concerns of many researchers is the recognition of sleep positions, which enables efficient monitoring of changes in sleeping posture for precise and intelligent adjustment. In sleep monitoring, machine learning is able to analyze the raw data collected and optimizes the algorithm in real-time to recognize the sleeping position of the human body during sleep.Methodology:A detailed search of relevant databases was conducted through a systematic search process, and we reviewed research published since 2017, focusing on 27 articles on sleep recognition.Results:Through the analysis and study of these articles, we propose several determinants that objectively affect sleeping posture recognition, including the acquisition of sleep posture data, data pre-processing, recognition algorithms, and validation analysis. Moreover, we analyze the categories of sleeping postures adapted to different body types.Conclusion:A systematic evaluation combining the above determinants provides solutions for system design and rational selection of recognition algorithms for sleep posture recognition, and it is necessary to regularize and standardize existing machine learning algorithms before they can be incorporated into clinical monitoring of sleep.