场景识别是普适计算中的一项重要研究内容,旨在通过识别智能手机用户所在位置的场景,为用户提供精准的个性化服务并提升服务的质量.在实际环境中,精确的场景识别存在两个问题:(1)基于单模传感器数据或无线信号数据的分类效果不佳、普适性不足;(2)场景识别的精度需要依赖大量标定数据,导致成本较高.针对这些问题,提出一种基于多模融合的半监督场景识别方法,该方法充分利用Wi-Fi、蓝牙和传感器的多模特征来提高识别精度.相比基于单模数据的识别,融合特征将静态场景的分类精度提升了10％,并且本文通过构建半监督的学习方法解决了动态场景中数据采集成本高的问题,在将标定数据量减少一半的基础上将识别精度提高至90％ 以上.实验数据表明,在利用Wi-Fi、蓝牙、传感器的互补优势的基础上,引入半监督的学习方法能够提升场景识别的精确度且降低在某些场景下采集数据的成本,从而有效地提升了场景识别的精度和普适性.

User's position and trajectory are very important personal informations, based on which, the user's personalized description can be realized, and the user's intentions, preferences and behavior patterns can also be found. For most of traditional approaches, the trajectory learning problems assume the availability of a vast amount of labeled training data, which requires a great deal of manual effort. In this paper we propose a indoor trajectory restoration method based on the spatial constraints on the position of points of interests (PoI). This method discovers PoI position by detecting the status changing of furniture and facilities and constructs spatial relationship of PoI positions with multidimensional scaling (MDS). Then, it synthetically optimizes the location model constructing of fitting error to labeled points, the manifold regularization of high-dimensional signal strength vectors and the MDS constraint of the spatial relationship of PoI positions, to achieve the optimizing trajectory by iteration. The experimental results show that our proposed method can effectively improve the robustness of the trajectory learning ability in different practical scenarios, benefiting from the constraint of PoI position.

室内定位是普适计算领域的热点研究问题.当前,室内定位方法主要分为基于信号传播模型的定位方法和基于无线信号指纹的定位方法.其中,基于指纹的方法由于不需要知道无线信号接入点(Access Point,AP)的位置而得到更加广泛的应用,其需要通过离线阶段采集大量数据来构建丰富的指纹库,满足这一条件需要大量的人工标定工作.对此,文中提出了一种基于指纹空间关系的定位方法,相比于传统的指纹定位方法,该方法无需建立指纹库,只需要通过获取多终端的Wi-Fi信号强度,计算所有终端的不相似度并构建不相似矩阵;通过多维尺度分析(Multidimensional Scaling,MDS)算法,构建出所有终端的位置分布图,进而通过确定其中3个以上终端的位置来定位所有的终端.采用支持向量回归机(Support Vector Regression,SVR)计算任意终端间的距离,并将距离矩阵作为不相似矩阵.文中在商场场景下选择了约2500m2的区域进行实验,所提方法的平均定位误差约为7m.

近年来室内位置服务的需求推进了多种室内定位技术的发展.目前的室内定位技术主要包括基于超声波、红外、RFID、超宽带、蓝牙、WLAN、地磁等载体途径的多种定位技术.各种定位技术具有不同的定位精度、应用载体和实施成本,可满足相应的场景定位需求.目前,商业地产、新零售、物流、智能制造、智慧城市等领域均出现大量应用案例.本文最后给出了北京客站定位导航系统的具体案例介绍.

Recently, the problem of indoor localization based on WLAN signals is attracting increasing attention due to the development of mobile devices and the widespread construction of networks. However, no definitive solution for achieving a low-cost and accurate positioning system has been found. In most traditional approaches, solving the indoor localization problem requires the availability of a large number of labeled training samples, the collection of which requires considerable manual effort. Previous research has not provided a means of simultaneously reducing human calibration effort and improving location accuracy. This paper introduces fusion semi-supervised extreme learning machine (FSELM), a novel semi-supervised learning algorithm based on the fusion of information from Wi-Fi and Bluetooth Low Energy (BLE) signals. Unlike previous semi-supervised methods, which consider multiple signals individually, FSELM fuses multiple signals into a unified model. When applied to sparsely calibrated localization problems, our proposed method is advantageous in three respects. First, it can dramatically reduce the human calibration effort required when using a semi-supervised learning framework. Second, it utilizes fused Wi-Fi and BLE fingerprints to markedly improve the location accuracy. Third, it inherits the beneficial properties of ELMs with regard to training and testing speeds because the input weights and biases of hidden nodes can be generated randomly. As demonstrated by experimental results obtained on practical indoor localization datasets, FSELM possesses a better semi-supervised manifold learning ability and achieves higher location accuracy than several previous batch supervised learning approaches (ELM, BP and SVM) and semi-supervised learning approaches (SELM, S-RVFL and FS-RVFL). Moreover, FSELM needs less training and testing time, making it easier to apply in practice. We conclude through experiments that FSELM yields good results when applied to a multi-signal-based semi-supervised learning problem. The contributions of this paper can be summarized as follows: First, the findings indicate that effective multi-data fusion can be achieved not only through data-layer fusion, feature-layer fusion and decision-layer fusion but also through the fusion of constraints within a model. Second, for semi-supervised learning problems, it is necessary to combine the advantages of different types of data by optimizing the model’s parameters.

随着WLAN的普及,基于RSS(Received Signal Strength)的室内定位方法逐渐成为研究与应用的热点.其中,基于指纹的定位方法已成为主流,此类方法的特点之一在于要求离线训练数据与在线测试数据满足独立同分布,然而,在实际环境中,现有的指纹定位方法或系统存在以下3个问题:1)不同终端设备的无线通讯硬件存在差异性,训练数据和测试数据的采集设备之间的差异性将严重影响定位精度;2)环境中的无线信号呈现高动态性,采集的数据存在显著的时效性,因此由训练数据得到的模型的定位性能将随着时间的推移不断下降;3)传统增量式定位模型需要大量的标定数据,不具有实际可用性.为解决以上问题,提出了一种针对设备差异性问题的增量式室内定位方法,利用终端在持续定位服务中采集的无标记数据来实时更新定位模型.实验表明,在实际蓝牙定位数据集上,相比于传统的定位模型方法,所提方法的整体定位精度更高,误差距离为3～5m时,其优势更为明显;同时,该方法具有时效优势,能够长时间保持有效定位.

在现代信息化生活中，人们对室内定位的需求日益增多，对系统定位精度要求也越来越高。如机场和大型购物广场内的导航、会议指南、资源查找、井下工作人员的定位、老人与儿童的监护等。本文针对室内定位在实际应用中的多方面问题进行了深入研究。具体工作包括：基于非标定信号流形特征的定位方法研究、设备与时间迁移的定位研究、大面积接入点（AP）无缝切换定位研究、垂直空间的相同楼宇定位模型迁移研究、面向接入点缺失与稀疏条件的定位研究、基于单接入点的房间级定位研究、免标定定位方法研究以及基于WiFi 定向特征的定位研究等。实验表明，基于WiFi 的室内定位技术日渐成熟，其定位精度以及系统的易用性正在使该技术向低成本实用化的目标迈进。

This paper presents a novel double knowledge transferring method to solve the multi-culture facial attractiveness enhancement problem. The existing enhancement of facial attractiveness methods just focus on one particular culture and assume the beautification model learned in one culture could be used in other cultures without adaptation. However, for the people in different cultures who do not share the same characteristics, existing models do not perform well. To address this problem, we employ the double knowledge transferring method to reduce data diversity. First, to reduce the aesthetic differences among cultures, we apply regressive manifold regularization to train a rating transferring function; second, to reduce the facial geometry feature differences, we use latent MLT (Multi-Task Learning) algorithm to find appropriate feature mappings. Treat Chinese as source culture, we apply the proposed algorithm to Malay people and demonstrate its effectiveness. Keyword: Facial attractiveness, transfer learning, manifold regularization.

A wearable auxiliary system was designed and implemented for the disability of people with autism in awareness of social proximity, which can help them to perceive, understand and respond to physical proximity with other people.A prototype system, ProCom was designed and developed, that combined self-build wearable device with a mobile phone application to support an individual in awareness of their real-time distance and orientation with interaction partners.Experimental results validated that ProCom was technically feasible and effective.ProCom system can not only support proximity awareness for people with autism in face-to-face social interactions, but also provide assistance to other researches on autism.

Wi-Fi and Bluetooth Low Energy based single mode localization methods cannot get satisfactory performance on localization accuracy, robustness and universality.In the training phase, a large amount of calibrated data is required to train a model.A semi-supervised localization method was proposed based on fusing features of Wi-Fi and Bluetooth low energy signals in order to solve these problems.Wi-Fi and Bluetooth Low Energy based localization methods were effectively used, and semi-supervised manifold was employed to import a vast amount of uncalibrated data for model training.Experimental results show that the proposed fusion feature can increase the indoor localization accuracy by more than 20% as well as improving robustness compared with single feature.The semi-supervised manifold localization method can dramatically reduce labeled calibration samples by 90%.

With the proliferation of healthcare data, the cloud mining technology for E-health services and applications has become a hot research topic. While on the other hand, these rapidly evolving cloud mining technologies and their deployment in healthcare systems also pose potential threats to patient’s data privacy. In order to solve the privacy problem in the cloud mining technique, this paper proposes a semi-supervised privacy-preserving clustering algorithm. By employing a small amount of supervised information, the method first learns a Large Margin Nearest Cluster metric using convex optimization. Then according to the trained metric, the method imposes multiplicative perturbation on the original data, which can change the distribution shape of the original data and thus protect the privacy information as well as ensuring high data usability. The experimental results on the brain fiber dataset provided by the 2009 PBC demonstrated that the proposed method could not only protect data privacy towards secure attacks, but improve the clustering purity.

As WLAN getting more and more popular and pervasive, Wi-Fi based indoor localization is becoming a hot issue in research and application fields. Among various kinds of up-to-date indoor localization methods, fingerprint based methods are most widely used because of the good performance. However, the existing fingerprint based methods still have following three common problems: Firstly, fingerprint based methods require a vast amount of calibration work, which need huge human and time consumption both in offline and online phases. It makes the systems difficult to be applied in the practical applications. Secondly, the Wi-Fi signals in the environment change frequently, bringing the significant timeliness in collected data. So it cannot guarantee to provide a long term effective localization. Thirdly, the Wi-Fi access points change frequently in real scene. Thus, the feature dimensions of training data and testing data are unequal. The traditional algorithms cannot well handle the feature dimension changing problem caused by increase or decrease in APs'number. To solve these problems mentioned above, we proposed a crowdsourcing based indoor localization method, including Semi-supervised ELM, Timeliness Managing ELM and Feature Adaptive Online Sequential ELM. We also developed an indoor localization platform. Applications show that our method can reduce human effort in data calibration and improve the model training speed. Moreover, our method can maintain the high location accuracy for a long time.

我们研发的“中国手语合成技术”能将任意给定的文本转换为手语并通过虚拟动画角色的多模态行为合成显示出来。本文在分析目前国内外手语合成技术和系统研究的基础上，重点介绍我们的工作，包括虚拟人建模、自然语言到手势语言转换、手势平滑、手势运动重定向以及动画角色多模态行为表达等关键技术，同时也介绍了基于中国手语合成技术的信息无障碍应用系统的研制，如：面向广电节目的电视手语制作系统和面向在线学习的奥运常用手语网络教学系统。最后介绍了目前还存在的问题以及下一步我们将开展的结合嵌入式终端和普适计算的研究工作。

Question subjectivity identification in Community Question Answering (CQA) has attracted a lot of attentions in recent years. With the rapid development of CQA, subjective questions posted by users are growing exponentially, which presents two challenges for question subjectivity identification. The first one is the data imbalance between subjective and objective questions. The second one is that the amount of manually labelled training data is hard to catch up with the fast developing speed of CQA. In this paper, we propose an adaptive semi-supervised Extreme Learning Machine (ASELM) to solve those two challenges. To resolve the data imbalance problem, ASELM employs the different impacts on identification performance caused by the imbalanced data. Second, the proposed method introduces the unlabelled data, and builds a model about the ratio between the number of labelled and unlabelled data based on Gaussian Model, which is applied to automatically generate the constraint on the unlabelled data. Experimental results showed ASELM improved identification performance for the imbalanced data, and outperformed the performance of basic ELM, SELM, Weighted ELM and SS-ELM on both F1 measure and accuracy.

In this paper, we describe the results of a controlled experiment measuring everyday movement activity through a novel recognition prototype named AIR. AIR measures distance from the feet using infrared (IR) sensors. We tested this approach for recognizing six prevalent activities: standing stationary, walking, running, walking in place, going upstairs, and going downstairs and compared results to other commonly used approaches. Our results show that AIR obtains much higher accuracy in recognizing activity than approaches that rely primarily on accelerometers. Moreover, AIR has good generalization ability when applying recognition model to new users.