Robust Vehicle Speed Measurement Based on Feature Information Fusion for Vehicle Multi-Characteristic Detection

A robust vehicle speed measurement system based on feature information fusion for vehicle multi-characteristic detection is proposed in this paper. A vehicle multi-characteristic dataset is constructed. With this dataset, seven CNN-based modern object detection algorithms are trained for vehicle multi-characteristic detection. The FPN-based YOLOv4 is selected as the best vehicle multi-characteristic detection algorithm, which applies feature information fusion of different scales with both rich high-level semantic information and detailed low-level location information. The YOLOv4 algorithm is improved by combing with the attention mechanism, in which the residual module in YOLOv4 is replaced by the ECA channel attention module with cross channel interaction. An improved ECA-YOLOv4 object detection algorithm based on both feature information fusion and cross channel interaction is proposed, which improves the performance of YOLOv4 for vehicle multi-characteristic detection and reduces the model parameter size and FLOPs as well. A multi-characteristic fused speed measurement system based on license plate, logo, and light is designed accordingly. The system performance is verified by experiments. The experimental results show that the speed measurement error rate of the proposed system meets the requirement of the China national standard GB/T 21555-2007 in which the speed measurement error rate should be less than 6%. The proposed system can efficiently enhance the vehicle speed measurement accuracy and effectively improve the vehicle speed measurement robustness.     

Radar sensor network for target detection using Chernoff information and relative entropy

In this paper, we propose to apply information theory to Ultra wide band (UWB) radar sensor network (RSN) to detect target in foliage environment. Information theoretic algorithms such as Maximum entropy method (MEM) and mutual information are proven methods, that can be applied to data collected by various sensors. However, the complexity of the environment poses uncertainty in fusion center. Chernoff information provides the best error exponent of detection in Bayesian environment. In this paper, we consider the target detection as binary hypothesis testing and use Chernoff information as sensor selection criterion, which significantly reduces the processing load. Another strong information theoretic algorithm, method of types, is applicable to our MEM based target detection algorithm as entropy is dependent on the empirical distribution only. Method of types analyzes the probability of a sequence based on empirical distribution. Based on this, we can find the bound on probability of detection. We also propose to use Relative entropy based processing in the fusion center based on method of types and Chernoff Stein Lemma. We study the required quantization level and number of nodes in gaining the best error exponent. The performance of the algorithms were evaluated, based on real world data.     

基于IFSR异常样本剔除的落叶松木材密度近红外优化模型的研究

木材密度可以反映木材的干缩性、抗压抗拉强度等多种物理性质,是重要的木材物理特性。采用近红外光谱技术能够实现木材密度的快速预测,可克服传统检测方法耗费人力、物力、时间的弊端,但建模结果往往受异常样本的影响。为准确识别并剔除样本集中的异常样本,提出一种孤立森林结合学生化残差方法(IFSR),在利用孤立森林集成特征的优点基础上考虑样本对模型的影响度,可同时检测异常样本与强影响样本。该研究对181个落叶松木材样本的近红外光谱及其在常温下的气干密度进行了测定。通过对比多种方法预处理和特征选择方法,确定采用标准正态变量变化(SNV)+去趋势处理(DT)+均值中心化(MC)+标准化(Auto)方法进行预处理,采用竞争性自适应重加权算法(CARS)进行特征波段选择,消除噪声及无关信息对算法的影响,简化数据集,提高算法剔除异常样本的准确性。为验证IFSR方法剔除异常样本的能力,将其与蒙特卡洛交互验证(MCCV)、马氏距离(MD)等其他六种异常检测方法对比分析,建立偏最小二乘(PLS)模型对其进行异常检测性能评价。同时在上述基础上采用粒子群寻优-支持向量机回归(PSO-SVR),BP神经网络(BPNN)与PLS分别建立落叶松木材密度近红外预测模型。结果表明,IFSR结合PSO-SVR方法得到的优化模型预测能力最强,IFSR可有效剔除奇异样本,提高模型精度。     

A novel approach for crowd video monitoring of subway platforms

Crowd monitoring in a dense crowd scene has become an important and challenging topic in the field of video surveillance system. This paper proposes a novel crowd monitoring approach for subway platforms to address requirements in rail traffic management. Firstly, an improvement for Mixture Gaussian background modeling is presented to segment the crowd. In the process of feature extraction, the concept of the weighted area is proposed to solve the problem of the perspective of images. To deal with the issue of the occlusion between individuals, an improved gradient feature is developed in this paper. And then, Adaptive Boost classifier with the feature weighted area and the improved gradient is used to estimate the crowd density. Finally, the crowd is counted by the method of linear regression. The experimental results show that the proposed approach is feasible and effective for crowd monitoring in real subway platforms.     

特征增强SSD算法及其在遥感目标检测中的应用

为了提高对复杂场景下多尺度遥感目标的检测精度,提出了基于多尺度单发射击检测(SSD)的特征增强目标检测算法.首先对SSD的金字塔特征层中的浅层网络设计浅层特征增强模块,以提高浅层网络对小目标物体的特征提取能力;然后设计深层特征融合模块,替换SSD金字塔特征层中的深层网络,提高深层网络的特征提取能力;最后将提取的图像特征与不同纵横比的候选框进行匹配以执行不同尺度遥感图像目标检测与定位.在光学遥感图像数据集上的实验结果表明,该算法能够适应不同背景下的遥感目标检测,有效地提高了复杂场景下的遥感目标的检测精度.此外,在拓展实验中,文中算法对图像中的模糊目标的检测效果也优于SSD.     

Contour level object detection with top-down information

This paper presents a contour level object detection approach. In contrast to conventional bounding box results, we give out the salient closed contour of the object, which provides a possibility of semantic analysis for the object. We get the salient closed contour with Ratio Contour algorithm. The top-down information needed by salient closed contour extraction is based on the well-known Bag-of-Features methodology. Our top-down information based contour extraction and completion is much more efficient and robust than many related approaches lack of the top-down information. We also propose a novel post-processing framework for object detection. With low threshold and a refined binary classifier, we can get stable high performance. We evaluate our approaches on UIUC cars dataset. We show that our approaches apparently improve the performance of object detections under clutter.     

基于中心点的遥感图像多方向舰船目标检测

针对目前基于深度学习的舰船目标斜框检测方法存在计算量大、效率低的问题,提出一种基于目标中心点的单阶段检测模型.由于舰船中心点不受舰船分布方向影响,模型主要思想是以目标中心点检测为基础,回归中心点处目标斜框的尺度和方向.首先设计特征提取网络,将卷积神经网络细节信息丰富的底层特征与语义信息丰富的高层特征融合起来形成特征图;然后将特征图输入到三个检测分支,分别预测目标中心点、中心点偏移值以及斜框的尺度与方向;设计组合损失函数对网络进行训练,并改进非极大值抑制算法以适应目标斜框检测的需要.在公开的SAR图像舰船目标检测数据集与光学遥感图像上进行了实验,实验结果表明,测试集平均准确率达0.906,检测精度与速度均优于其它检测模型,充分验证了所提算法的有效性.     

Clustering gene expression data analysis using an improved EM algorithm based on multivariate elliptical contoured mixture models

Clustering gene expression data is an important research topic in bioinformatics because knowing which genes act similarly can lead to the discovery of important biological information. Many clustering algorithms have been used in the field of gene clustering. The multivariate Gaussian mixture distribution function was frequently used as the component of the finite mixture model for clustering, however the clustering cannot be restricted to the normal distribution in the real dataset. In order to make the cluster algorithm strong adaptability, this paper proposes a new scheme for clustering gene expression data based on the multivariate elliptical contoured mixture models (MECMMs). To solve the problem of over-reliance on the initialization, we propose an improved expectation maximization (EM) algorithm by adding and deleting initial value for the classical EM algorithm, and the number of clusters can be treated as a known parameter and inferred with the QAIC criterion. The improved EM algorithm based on the MECMMs is tested and compared with some other clustering algorithms, the performance of our clustering algorithm has been extensively compared over several simulated and real gene expression datasets. Our results indicated that improved EM clustering algorithm is superior to the classical EM algorithm and the support vector machines (SVMs) algorithm, and can be widely used for gene clustering.     

Experimental study and numerical simulation of evacuation from a dormitory

The evacuation process of students from a dormitory is investigated by both experiment and modeling. We investigate the video record of pedestrian movement in a dormitory, and find some typical characteristics of evacuation, including continuous pedestrian flow, mass behavior and so on. Based on the experimental observation, we found that simulation results considering pre-movement time are closer to the experimental results. With the model considering pre-movement time, we simulate the evacuation process and compare the simulation results with the experimental results, and find that they agree with each other closely. The crowd massing phenomenon is conducted in this paper. It is found that different crowd massing phenomena will emerge due to different desired velocities. The crowd massing phenomenon could be more serious with the increase of the desired velocity. In this study, we also found the faster-is-slower effect. When the positive effect produced by increasing the desired velocity is not sufficient for making up for its negative effect, the phenomenon of the greater the desired velocity the longer the time required for evacuation will emerge. From the video record, it can be observed that the mass behavior is obvious during the evacuation process. And the mass phenomenon could also be found in simulation. The results obtained from our study are also suitable to all these buildings in which both living and resting areas occupy the majority space, such as dormitories, residential buildings, hotels (restaurants) and so on.     

基于自适应阈值和动态更新因子的ViBe改进算法

针对传统ViBe算法不能及时反映场景变化,动态场景适应性差等问题,提出一种改进的ViBe算法。改进内容包括:采用随机选取背景样本和24邻域法获取初始背景,可以加速“鬼影”消融;结合大津法（OTSU）和均匀性度量法的平均自适应阈值计算方法,可以提高算法对树叶晃动、水波纹和光照变化等环境的适应性,最大限度保留有效像素;更新阶段引入自适应更新因子,可以有效减少被误判的概率,从而增强算法的鲁棒性;最后通过形态学处理和滤波使目标更加完整。采用标准数据集视频对改进算法进行了测试和对比分析,改进算法相对于KDE算法、GMM算法和传统ViBe算法各项指标均有大幅度提高,精确度分别提高30.44%、40.72%和20.95%,错分比分别降低了43.28%、40.59%和29.43%。     

Pedestrian intention prediction based on dynamic fuzzy automata for vehicle driving at nighttime

In this paper, we propose a novel algorithm that can predict a pedestrian’s intention using images captured by a far-infrared thermal camera mounted on a moving car at nighttime. To predict a pedestrian’s intention in consecutive sequences, we use the dynamic fuzzy automata (DFA) method, which not only provides a systemic approach for handling uncertainty but also is able to handle continuous spaces. As the spatio-temporal features, the distance between the curbs and the pedestrian and the pedestrian’s velocity and head orientation are used. In this study, we define four intention states of the pedestrian: Standing-Sidewalk (S-SW), Walking-Sidewalk (W-SW), Walking-Crossing (W-Cro), and Running-Crossing (R-Cro). In every frame, the proposed system determines the final intention of the pedestrian as ‘Stop’ if the pedestrian’s intention state is S-SW or W-SW. In contrast, the proposed system determines the final intention of a pedestrian as ‘Cross’ if the pedestrian’s intention state is W-Cro or R-Cro. A performance comparison with other related methods shows that the performance of the proposed algorithm is better than that of other related methods. The proposed algorithm was successfully applied to our dataset, which includes complex environments with many pedestrians.     

基于目标与环境FD模型的多特征检测算法适应性评估

智能变形、变色、变温、变谱技术发展趋势下,低特征目标加速实现与自然地物背景的特征融合,导致复杂自然背景环境下低散射、微反射、弱辐射目标的检测与评估愈发困难,特定场景下潜在威胁目标的检测方法快速决策与准确评估成为了难题。为了提升离散目标、伪装目标、弱小目标、异常目标等低特征目标与复杂自然背景环境融合场景下的多特征检测算法的选择效率及其检测准确度,提出了目标与背景环境融合度（FD）参数模型,并设计了植被伪装目标嵌入草地背景、植被伪装目标嵌入土壤背景、植被及水泥路伪装目标嵌入土壤背景以及植被、水泥路、土壤伪装目标分别嵌入草地、水泥路、土壤背景等4种不同波谱特征分布场景的模拟图像数据,以及信噪比为200,400与800的高斯白噪声分别加入场景一的3种不同级别噪声比例的模拟图像数据。通过综合目标波谱信息、背景波谱信息、数据噪声比例等多种因素的综合试验分析,开展了基于目标与环境FD模型的多特征检测算法适应性评估研究。结果表明,在标准差均小于0.08的条件下,MtACE,MtAMF,MtCEM,SumACE,SumAMF,SumCEM,WtaACE,WtaAMF,WtaCEM等9大经典多特征检测算法对于4种波谱分布场景检测结果的FD参数平均值分别为0.320 0,0.350 2,0.862 4,0.365 8,0.365 8,0.846 1,0.680 0,0.680 0和0.948 2;在标准差均小于0.07的条件下,9大经典多特征检测算法对于3种不同级别噪声比例数据检测结果的FD参数平均值分别为0.313 5,0.320 9,0.774 7,0.369 6,0.369 6,0.847 5,0.695 6,0.695 6和0.960 3。通过不同波谱分布场景及不同噪声级别条件下的检测与融合度评估试验分析,实现了多特征检测算法的适应性能排序,大幅提升复杂场景下多种低特征目标的检测效率。综合波谱与噪声因素,对于复杂场景下离散分布的低特征目标检测,9大经典多特征检测算法的优先级顺序为：MtACE>MtAMF>SumACE=SumAMF>>WtaACE=WtaAMF>MtCEM>SumCEM>WtaCEM。     

Incorporating topography in a cellular automata model to simulate residents evacuation in a mountain area in China

In China, both the mountainous areas and the number of people who live in mountain areas occupy a significant proportion. When production accidents or natural disasters happen, the residents in mountain areas should be evacuated and the evacuation is of obvious importance to public safety. But it is a pity that there are few studies on safety evacuation in rough terrain. The particularity of the complex terrain in mountain areas, however, makes it difficult to study pedestrian evacuation. In this paper, a three-dimensional surface cellular automata model is proposed to numerically simulate the real time dynamic evacuation of residents. The model takes into account topographic characteristics (the slope gradient) of the environment and the biomechanics characteristics (weight and leg extensor power) of the residents to calculate the walking speed. This paper only focuses on the influence of topography and the physiological parameters are defined as constants according to a statistical report. Velocity varies with the topography. In order to simulate the behavior of a crowd with varying movement velocities, and a numerical algorithm is used to determine the time step of iteration. By doing so, a numerical simulation can be conducted in a 3D surface CA model. Moreover, considering residents evacuation around a gas well in a mountain area as a case, a visualization system for a three-dimensional simulation of pedestrian evacuation is developed. In the simulation process, population behaviors of congestion, queuing and collision avoidance can be observed. The simulation results are explained reasonably. Therefore, the model presented in this paper can realize a 3D dynamic simulation of pedestrian evacuation vividly in complex terrain and predict the evacuation procedure and evacuation time required, which can supply some valuable information for emergency management.     

Neural Estimator of Information for Time-Series Data with Dependency

Novel approaches to estimate information measures using neural networks are well-celebrated in recent years both in the information theory and machine learning communities. These neural-based estimators are shown to converge to the true values when estimating mutual information and conditional mutual information using independent samples. However, if the samples in the dataset are not independent, the consistency of these estimators requires further investigation. This is of particular interest for a more complex measure such as the directed information, which is pivotal in characterizing causality and is meaningful over time-dependent variables. The extension of the convergence proof for such cases is not trivial and demands further assumptions on the data. In this paper, we show that our neural estimator for conditional mutual information is consistent when the dataset is generated with samples of a stationary and ergodic source. In other words, we show that our information estimator using neural networks converges asymptotically to the true value with probability one. Besides universal functional approximation of neural networks, a core lemma to show the convergence is Birkhoff’s ergodic theorem. Additionally, we use the technique to estimate directed information and demonstrate the effectiveness of our approach in simulations.     

基于高斯混合模型的语音带宽扩展算法的研究

为了降低高带谱失真,研究了带宽扩展算法中特征参数与高带谱包络的互信息和高带谱失真之间的函数关系,并在此基础上提出了一种扩展高斯混合模型带宽扩展算法。首先,算法选择与高带谱包络互信息大的参数构成特征矢量,并根据高斯混合模型计算特征矢量与高带谱包络的联合概率密度。其次,采用Expectation-Maximization(EM)算法估计高斯分量模型参数并计算后验概率。最后,通过后验概率估计高带谱包络。实验结果表明,与传统的高斯混合模型带宽扩展算法相比,本文算法可降低0.3 dB的高带平均谱失真,将谱失真大于10dB的语音帧减少了50%以上。      

Research on Community Detection in Complex Networks Based on Internode Attraction

With the rapid development of computer technology, the research on complex networks has attracted more and more attention. At present, the research directions of cloud computing, big data, internet of vehicles, and distributed systems with very high attention are all based on complex networks. Community structure detection is a very important and meaningful research hotspot in complex networks. It is a difficult task to quickly and accurately divide the community structure and run it on large-scale networks. In this paper, we put forward a new community detection approach based on internode attraction, named IACD. This algorithm starts from the perspective of the important nodes of the complex network and refers to the gravitational relationship between two objects in physics to represent the forces between nodes in the network dataset, and then perform community detection. Through experiments on a large number of real-world datasets and synthetic networks, it is shown that the IACD algorithm can quickly and accurately divide the community structure, and it is superior to some classic algorithms and recently proposed algorithms.     

The moving target detection algorithm based on the improved visual background extraction

This paper presents a moving target detection algorithm based on the improved visual background extraction. Traditional VIBE (Visual Background Extractor) algorithm is one of the powerful background subtraction algorithm. It can quickly, accurately and integrally detect moving target. However, sometimes it will falsely determine background as foreground and impact detection results. In this paper, we improve the traditional VIBE algorithm by joining TOM (Time of map) mechanism in the process of detection, so it can not only use the pixel’s spatial domain information, but also make full use of the pixel’s time domain information. Experiments detailed in this paper show the algorithm presented in this paper has better detection effect than the traditional VIBE algorithm.     

Estimating and adjusting abnormal networks with unknown parameters and topology

The changes of parameters and topology in a complex network often lead to unexpected accidents in complex systems, such as diseases in neural systems and unexpected current in circuit system, so the methods of adjusting the abnormal network back to its normal conditions are necessary to avoid these problems. However, it is not easy to detect the structures and information of each network, even if we can find a network which has the same function as the abnormal network, it is still hard to use it as a reference to adjust the abnormal network because a lot of network information is unknown. In this paper, we design a "bridging network" as an information bridge between a normal network and an abnormal network to estimate and control the abnormal network. Through the "bridging network" and some adaptive laws, the abnormal parameters and connections in abnormal network can be adjusted to the same conditions as those of the normal network which is chosen as a reference model. Finally, the "bridging network" and the abnormal network achieve synchronization with the normal network. Besides, the detailed inner information in normal network and abnormal network can be accurately estimated by this "bridging network." Finally, the nodes in the abnormal network will behave normally after the correction. In this paper, we use Hindmarsh-Rose model as an example to describe our method.     

Preliminary abnormal electrocardiogram segment screening method for Holter data based on long short-term memory networks

Holter usually monitors electrocardiogram(ECG)signals for more than 24 hours to capture short-lived cardiac abnormalities.In view of the large amount of Holter data and the fact that the normal part accounts for the majority,it is reasonable to design an algorithm that can automatically eliminate normal data segments as much as possible without missing any abnormal data segments,and then take the left segments to the doctors or the computer programs for further diagnosis.In this paper,we propose a preliminary abnormal segment screening method for Holter data.Based on long short-term memory(LSTM)networks,the prediction model is established and trained with the normal data of a monitored object.Then,on the basis of kernel density estimation,we learn the distribution law of prediction errors after applying the trained LSTM model to the regular data.Based on these,the preliminary abnormal ECG segment screening analysis is carried out without R wave detection.Experiments on the MIT-BIH arrhythmia database show that,under the condition of ensuring that no abnormal point is missed,53.89% of normal segments can be effectively obviated.This work can greatly reduce the workload of subsequent further processing.