基于BING似物性检测的行人快速检测算法*

针对传统滑动窗行人检测速度慢、实时性差的问题,提出了一种基于似物性的行人快速检测算法。首先,算法通过提取正负训练样本的规范化二进制梯度特征,训练级联SVM分类器得到行人似物检测模型。然后利用尺寸调节和聚类算法对初始候选区域进行聚类融合,进一步优化行人候选窗口区域。最后,提取各候选区域的HOG特征并利用SVM分类器对其进行进一步行人检测。实验结果表明：本算法在保证行人检测率的同时在检测实时性上有明显提高。     

基于似物性的快速视觉目标识别算法

针对视觉目标识别算法实时性较差的问题,基于似物性提出一种面向视觉目标识别的可变部件模型改进算法。该算法首先对图像进行二进制归一化的似物性检测,并利用检测结果形成视觉目标候选框;然后使用目标识别算法对候选区域进行似然判决,比滑动窗口法缩短了搜索时间;最后通过一个快速扩大-缩小算法对检测目标进行尺度修正,提高目标框的准确度。在PASCAL 图像库上的识别结果表明:该识别方法在准确率上优于当前主流的检测模型,计算耗时较级联DPM算法减少约50%。     

EPMDroid: Efficient and privacy-preserving malware detection based on SGX through data fusion

Android has stood at a predominant position in mobile operating systems for many years. However, its popularity and openness make it a desirable target of malicious attackers. There is an increasing need for mobile malware detection. Existing analysis methods fall into two categories, i.e., static analysis and dynamic analysis. The dynamic analysis is more effective and timely than the static one, but it incurs a high computational overhead, thus cannot be deployed in resource-constrained mobile devices. Existing studies solve this issue by outsourcing malware detection to the cloud. However, the privacy of mobile app runtime data uploaded to the cloud is not well preserved during both detection model training and malware detection. Numerous efforts have been made to preserve privacy with cryptography, which suffers from high computational overhead and low flexibility. To address these issues, in this paper, we propose an Intel SGX-empowered mobile malware detection scheme called EPMDroid. We also design a probabilistic data structure based on cuckoo filters, named CuckooTable, to effectively fuse features for detection and achieve high space efficiency. We conduct both theoretical analysis and real-world data based tests on EPMDroid performance. Experimental results show that EPMDroid can speed up malware detection by up to 43.8 times and save memory space by up to 3.7 times with the same accuracy, as compared to a baseline method.     

基于轻量级网络的实时人体关键点检测算法

为提升人体姿态估计在移动终端设备上的运行速度与实时性,提出一种改进的人体关键点检测算法。通过将MobileNetV2轻量级主干网络与深度可分离卷积模块相结合加速特征提取过程,使用精炼网络进行多尺度人体关键点预测,并利用融合网络整合多个尺度的预测结果得到最终人体关键点检测结果。实验结果表明,与传统CPM算法相比,该算法在网络模型参数量和浮点运算量明显减少的情况下PCKh@0.5仅下降0.1个百分点,具有较高的检测精度和较好的实时性。     

结合网格密度聚类的行人检测候选域生成

目的 行人检测是计算机视觉领域中的重点研究问题。经典的可变形部件模型（DPM）算法在行人检测领域素有高检测精度的优点,但由于在构建特征金字塔前处理过多召回率低的候选区域,导致计算速度偏慢,严重影响系统的实时性。针对该问题,本文对模型中选取候选检测区域的流程进行了改进,提出一种结合网格密度聚类算法和选择性搜索算法的行人检测候选对象生成方法来改进DPM模型。方法 首先使用三帧差法和高斯混合模型收集固定数量的运动物体坐标点,然后结合基于网格密度的聚类算法构建网格坐标模型,生成目标频繁运动区域,同时进行动态掩层处理。随后引入改进的选择性搜索算法,结合支持向量机（SVM）训练得到的行人轮廓宽高比,提取该区域中高置信度的行人候选检测窗口,从而排除大量冗余的区域假设,完成对候选行人检测区域的精筛选,最后融合至DPM算法进行行人检测。结果 所提方法在PETS 2009 Bench-mark数据集上进行检测,实验结果表明,该方法对复杂背景下的检测有较强的稳定性,与传统DPM模型相比,精度提高了1.71%、平均对数漏检率降低2.2%、检测速度提高为3.7倍左右。结论 本文提出一种基于网格密度聚类的行人检测候选域生成算法,能够有效表达行人信息,与其他行人检测算法相比,有更好的精度和更快的速度,在检测率、检测时间方面均有提高,能够实现有效、快速的行人检测,具有实际意义。     

Context inclusive function evaluation: a case study with EM-based multi-scale multi-granular image classification

Many statistical queries such as maximum likelihood estimation involve finding the best candidate model given a set of candidate models and a quality estimation function. This problem is common in important applications like land-use classification at multiple spatial resolutions from remote sensing raster data. Such a problem is computationally challenging due to the significant computation cost to evaluate the quality estimation function for each candidate model. For example, a recently proposed method of multi-scale, multi-granular classification has high computational overhead of function evaluation for various candidate models independently before comparison. In contrast, we propose an upper bound based context-inclusive approach that reduces computational overhead based on the context, i.e. the value of the quality estimation function for the best candidate model so far. We also prove that an upper bound exists for each candidate model and the proposed algorithm is correct. Experimental results using land-use classification at multiple spatial resolutions from satellite imagery show that the proposed approach reduces the computational cost significantly.     

Multistage particle windows for fast and accurate object detection

The common paradigm employed for object detection is the sliding window (SW) search. This approach generates grid-distributed patches, at all possible positions and sizes, which are evaluated by a binary classifier: The tradeoff between computational burden and detection accuracy is the real critical point of sliding windows; several methods have been proposed to speed up the search such as adding complementary features. We propose a paradigm that differs from any previous approach since it casts object detection into a statistical-based search using a Monte Carlo sampling for estimating the likelihood density function with Gaussian kernels. The estimation relies on a multistage strategy where the proposal distribution is progressively refined by taking into account the feedback of the classifiers. The method can be easily plugged into a Bayesian-recursive framework to exploit the temporal coherency of the target objects in videos. Several tests on pedestrian and face detection, both on images and videos, with different types of classifiers (cascade of boosted classifiers, soft cascades, and SVM) and features (covariance matrices, Haar-like features, integral channel features, and histogram of oriented gradients) demonstrate that the proposed method provides higher detection rates and accuracy as well as a lower computational burden w.r.t. sliding window detection.     

A generic partial encryption scheme for low-power mobile devices

Content protection that allows only legitimate users to use specified content is essential in order to secure business in the consumer market. However, service providers and users suffer from low responsiveness when content is encrypted with traditional cryptographic tools that require strong decryption algorithms on mobile devices. In this paper, we introduce a generic partial encryption scheme for low-power mobile devices. Our primary goal is to design a generic architecture for partial encryption of downloadable and real-time streaming contents, and also to facilitate a trade-off between minimizing the encryption/decryption overhead and providing sufficient DRM security for the service provider. We also evaluate the efficacy of our proposed scheme by applying it to real-world multimedia contents. The results of our experiments indicate that encrypting only a small portion (about 2.5 %) of video content can effectively impose DRM restriction on the content. This significantly reduces the decryption overhead on low-power mobile devices. In the smart phone environment, it is shown that the time overhead during the decryption is less that 5 % of on-the-fly decoding time and the power overhead is reduced by up to 94.5 %, compared to the traditional full encryption scheme.     

改进星型级联可形变部件模型的行人检测

目的 行人检测是计算机视觉和模式识别领域的研究热点与难点,由于经典的可形变部件模型(DPM)检测速度太慢,引入PCA降维的星型级联检测可形变部件模型(casDPM)相比较于DPM模型检测速度虽然有了很大提升,但在应用于行人检测时,出现检测精度较低、平均对数漏检率较高的情况,为了更加准确地对行人进行检测,提出了一种改进casDPM模型的行人检测方法。方法 首先利用对象度量方法获取目标候选区域,结合目标得分信息得到casDPM模型低分检测区域的置信度,在设定的阈值上保留检测窗口;然后针对casDPM模型原有非极大值抑制(Nms)算法只利用单一的面积信息,造成误检数较高的情况,提出了利用检测窗口的得分信息进行改进;最后将两种方法结合起来,提出了融合的cas-WNms-BING模型。结果 采用本文方法在INRIA数据集上进行检测,实验结果表明该方法对于行人形变、背景特征复杂及遮挡现象具有较强的鲁棒性,相比casDPM模型,本文提出的方法平均精度(AP)可以提高1.74%,平均对数漏检率可以降低4.45%。结论 提出一种改进星型级联可形变部件模型,取得一定的研究成果,在复杂的背景下,能够有效地进行行人检测,主观视觉感受和客观实验评价指标都表明该方法可以有效提升模型行人检测效果。但是,星型级联可形变部件模型训练及检测效率仍有待提高,需进一步对模型存在的一些局限性进行深入研究。     

面向移动通信网的蓝牙加密语音同步研究

在蓝牙终端嵌入加密模块，可以实现语音在手机和移动通信网的安全传输，针对蓝牙加密语音同步问题，通过正弦信号时域特征提取，利用波形码本的可透传性，设计了初始同步头和周期同步头，提出了语音初始同步建立方案和语音同步检测恢复方案。仿真实验及分析表明，方案具有较小时间开销和计算开销，能够有效实现初始同步、同步检测及恢复。     

光流估计下的移动端实时人脸检测

为了提高移动设备人脸检测准确率,提出一种应用于移动设备的实时人脸检测算法。通过改进Viola-Jones方法进行人脸区域快速分割,在不损失速度的情况下提高分割精度;同时应用了光流估计方法将卷积神经网络子网络在离散关键帧上的特征提取结果传播至非关键帧,提高神经网络实际检测运行效率。实验使用YouTube视频人脸数据库、自建20人各1 min正位人脸视频数据库和实际检测项目在不同分辨率下进行,实验结果表明运行速度在2.35帧/秒~22.25帧/秒,达到了一般人脸检测水平;人脸检测在10%误检率下召回率由Viola-Jones的65.93%提高到82.5%~90.8%,接近卷积神经网络检测精度,满足了移动设备实时人脸检测的速度和精度要求。     

基于似物性和空时协方差特征的行人检测算法

针对行人检测算法中缺少空时信息融合、检测区域过大等问题,提出了一种联合似物性检测和基于通道协方差信息的改进算法。该算法首先对图像进行二进制梯度归一化的似物性检测,并形成行人检测候选区域,缩小检测区域;然后提取待测目标的空间和时间特征;最后基于协方差信息构造一种融合空时特征的检测器,以提高检测精度。在公开的数据集INRIA和Caltech上的实验结果表明:该算法的性能优于目前主流的行人检测算法。     

可验证医疗密态数据聚合与统计分析方案

随着移动通信网络的飞速发展,越来越多的可穿戴设备通过移动终端接入网络并上传大量医疗数据,这些医疗数据聚合后具有重要的医学统计分析与决策价值.然而,在医疗数据传输和聚合过程中会出现传输中断、信息泄露、数据篡改等问题.为了解决这些安全与隐私问题,同时支持高效而正确的医疗密态数据聚合与统计分析功能,提出了基于移动边缘服务计算的具有容错机制的可验证医疗密态数据聚合方案.该方案改进了BGN同态加密算法,并结合Shamir秘密共享机制,确保医疗数据机密性、密态数据的可容错聚合.该方案提出了移动边缘服务计算辅助无线体域网的概念,结合移动边缘计算和云计算,实现海量医疗大数据实时处理与统计分析.该方案通过边缘计算服务器和云服务器两层聚合,提高聚合效率,降低通信开销.同时,使用聚合签名技术实现医疗密态数据的批量验证功能,进而保障其在传输与存储过程中的完整性.性能比较与分析表明,该方案在计算与通信开销方面都具备突出优势.     

Learning rich features from objectness estimation for human lying-pose detection

Lying-pose human detection is an active research field of computer vision in recent years. It has a good theoretical significance and furthermore many applications, such as victim detection or home service robot. But the study on lying-pose human detection in low-altitude overlooking images have many unsolved problems owing to multiple poses, arbitrary orientation, in-plane rotation, perspective distortion, and time-consuming. In this paper, the proposed framework of human lying-pose detection is optimization and machine learning algorithms inspired by processes of neurobiology suggest and human vision system to select possible object locations. First, the proposed model effectively utilizes binarized normed gradient features to obtain the objectness rapidly based on the vision saliency. Further, deep-learning techniques based on the convolution neural network are trained for learning rich feature hierarchies, in order to obtain the object of lying-pose human from objectness estimation, unlike the classical sliding-window algorithm. Eventually, employed pyramid mean-shift algorithm and rotation-angle recovery method to find position and direction of human lying-pose. The experimental results show that our method is rapid and efficient, and that it achieves state-of-the-art results with our XMULP dataset.     

Real-time collaborative reconstruction of digital building models with mobile devices

Many software-based building processes require digital building models. Since the building stock does not have sufficient data in this regard, the demand for Scan-to-BIM processes is increasing. In this paper we present a system for the reconstruction of ‘as-built’ BIM content of house interiors based on the Google Tango technology. The strength of our approach is the use of low-cost mobile scanning devices and a client-server system that allows for a real-time collaborative scanning and reconstruction of indoor scenes. We developed a server application that continuously aggregates scan data of multiple scanning devices (clients) and applies the data stream to a real-time post-processing pipeline to reconstruct rooms, walls, doors and windows. The reconstruction result is then distributed to all clients, where it is visualized in real time. The collaborative workflow and real-time data processing make our system especially useful in situations that are time-critical and require concurrent collection and processing of data. One of our targeted use cases therefore is the model generation for crime scene documentation. The effectiveness of our approach was demonstrated on three test sites. Our results compare well to other state-of-art methods regarding the reconstruction of walls, but they also revealed potential for improvement regarding the detection of doors and windows in occluded and cluttered environments.     

Enhanced MAC protocol to support multimedia traffic in cognitive wireless mesh networks

With the advanced physical layer techniques such as multiple-input and multiple-output (MIMO) and orthogonal frequency-division multiplexing (OFDM), transmission real-time 2D/3D contents and applications becomes more and more necessary in wireless networks for the amazing growing in demand of customers. However, the low efficiency of medium access control (MAC) protocol degrades the performance of real-time traffic greatly in multihop, wireless and mobile environment. Focusing on supporting real-time multimedia traffic in cognitive wireless mesh networks (WMNs), an enhanced MAC protocol is proposed. And the contribution of this paper is twofold: (1) An efficient carrier sense multiple access with collision avoidance (CSMA/CA) compatible time division multiple access (TDMA)-like MAC protocol called T-MAC is proposed, which aims to improve the system performance by allocating more channel access time in centralized manner and decreasing overhead. (2) An optimal adaptive scheduling scheme is proposed to support real-time multimedia applications and guarantee QoS for different priority traffic, which aims to find the optimized schedule among all possible sequences of concurrent transmissions by minimizing the occupied resources. Detailed simulation results and comparison with IEEE 802.11e MAC scheme show that the proposed T-MAC can effectively improve quality of service (QoS) for multimedia traffic in terms of throughput, end-to-end delay and packet loss rate, which also manifests that T-MAC is an efficient multimedia applications transmission scheme for mobile terminals and MAPs in cognitive WMNs.     

Compact Differential Evolution Light: High Performance Despite Limited Memory Requirement and Modest Computational Overhead

Compact algorithms are Estimation of Distribution Algorithms which mimic the behavior of population-based algorithms by means of a probabilistic representation of the population of candidate solutions. These algorithms have a similar behaviour with respect to population-based algorithms but require a much smaller memory. This feature is crucially important in some engineering applications, especially in robotics. A high performance compact algorithm is the compact Differential Evolution (cDE) algorithm. This paper proposes a novel implementation of cDE, namely compact Differential Evolution light (cDElight), to address not only the memory saving necessities but also real-time requirements. cDElight employs two novel algorithmic modifications for employing a smaller computational overhead without a performance loss, with respect to cDE. Numerical results, carried out on a broad set of test problems, show that cDElight, despite its minimal hardware requirements, does not deteriorate the performance of cDE and thus is competitive with other memory saving and population-based algorithms. An application in the field of mobile robotics highlights the usability and advantages of the proposed approach.     

CAD data visualization on mobile devices using sequential constrained Delaunay triangulation

3D graphic rendering in mobile application programs is becoming increasingly popular with rapid advances in mobile device technology. Current 3D graphic rendering engines for mobile devices do not provide triangulation capabilities for surfaces; therefore, mobile 3D graphic applications have been dealing only with pre-tessellated geometric data. Since triangulation is comparatively expensive in terms of computation, real-time tessellation cannot be easily implemented on mobile devices with limited resources. No research has yet been reported on real-time triangulation on mobile devices.In this paper, we propose a real-time triangulation algorithm for visualization on mobile devices based on sequential constrained Delaunay triangulation. We apply a compact data structure and a sequential triangulation process for visualization of CAD data on mobile devices. In order to achieve a high performance and compact implementation of the triangulation, the nature of the CAD data is fully considered in the computational process. This paper also presents a prototype implementation for a mobile 3D CAD viewer running on a handheld Personal Digital Assistant (PDA).     

TEASE: A novel Tunnel-based sEcure Authentication SchemE to support smooth handoff in IEEE 802.11 wireless networks

With the growing popularity of WiFi-based devices, WiFi-based wireless networks have received a great deal of interest in the wireless networks community. However, due to the limited transmission range of WiFi-based networks, mobile users have to switch their associated access points constantly to maintain continuing communications during their movement. The process of switching access points is called handoff. Handoff management is a key service in mobile networks, because providing seamless roaming in wireless networks is mandatory for supporting real-time applications in a mobile environment, such as VoIP, online games, and eConference. Security is another important issue in network communications, and to prevent possible attacks, authentication is required during the handoff process to guarantee the reliability of mobile clients and access points. In this paper, we propose a novel authentication scheme to achieve a smooth handoff in WiFi-based networks, which we refer to as TEASE. A tunnel is introduced to forward data packets between the new access point and the original reliable access point. The processing of a complete secure authentication and the transmitting of data between mobile terminals and their correspondence nodes can go on simultaneously. The security of handoff is achieved without increasing overhead to authentication servers, and handoff latency can be minimized to support seamless roaming. Simulation results show that our proposed scheme reduces significantly the communication interruption time and generates low packet loss ratio, and our method is suitable to be used for secure handoff in real-time applications.     

LTE系统基于格基约减辅助V-BLAST算法

LTE作为以OFDM-MIMO为主要技术特征的第四代移动通信,它的终端信号检测实现比较困难,这就需要一种性能好、复杂度低的检测算法来实现。格基约减是一种在接收端对信道矩阵进行预处理,可以消除子信道间干扰和抑制噪声的增强。本文在已有的格基约减ELLL算法的基础上,提出一种限制条件更为宽松的对角格约减算法(DR)。该算法的计算复杂度要低于ELLL算法。在该算法的基础上,结合传统V-BLAST和K-best算法思想,给出了一种基于格基约减辅助的V-BLAST算法。仿真结果表明,在LTE系统中该算法能够在复杂度较低的情况下,性能更接近ML算法。