Abstract:Because human face tracking is prone to be affected under complex postures (largely tilt, rotate, occlusion and color interference), this paper proposes a new face tracking algorithm based on online modifying, which includes two modules: detection module and tracking module. Tracking module is effective when detection is failed, and in the process of face tracking, in order to reduce the continuous accumulation of errors, we use real-time face detection to fix the parameters (including tracking the location and scale) of the tracking module, which made use of face detection and tracking with their respective advantages. Experiments show that the algorithm is able to accurately track a face under largely tilt and rotate and is also robust to color interfere and occlusion. In addition, the algorithm is used to control commercial games, which provides a new interactive way for HCI.

Abstract:The storage and query of context is the foundation of context awareness based computing. At the same time, uncertainty of context is an inevitable problem of the context awareness. Therefore, a fuzzy logic embedded context-aware architecture (FLECA) is proposed in this paper. FLECA gets the high level context semantic from querying those primary contexts. During the procedure of management of contexts, fuzzy logic engine enables FLECA to store and query the fuzzificated contexts. Furthermore, a neural network based learning machine is introduced into the management of fuzzy rules. The experiment shows that the learning machine is able to improve the adaptability of FLECA to provide stronger support to the user-centric applications.

Abstract:Recognition of movements during sleeping is the foundation of sleep quality assessment. This paper proposes an unobtrusive system for activity detection in bed, which collects data by using force sensors, develops user’s individualized model according to the theory of statistical pattern recognition, and identifies the user’s activity through algorithm of maximum similarity at last. This system does not interfere with the normal life and is easy to deploy. The prototype was implemented and the results showed that the recognition accuracy of sleeping behaviors is above 80%, which illustrates the feasibility and practical value of the system.

Abstract:Interactive tabletop is now receiving wide publicity for its natural setting for collaboration. However, few works have carefully studied the user interface management on tabletop, and many problems are still to be resolved. In this paper, we carry out a systematic research on tabletop UI management, and develop a system to facilitate user interaction on tabletop. Firstly, based on the analysis of tabletop’s horizontal display, multi-user usage and natural input features, we identify four fundamental requirements for tabletop UI management: input management, layout management, conflict management and authority management, which are significantly different from UI management on desktops. Secondly, we design a model for tabletop UI management. It contains a hierarchical structure for organizing surface elements, Object Manager for multi-user interface controlling, and Input Manager for processing concurrent multimodal inputs. Thirdly, we implement a system, uTablePlatform based on our model, and an SDK for users developing applications on uTablePlatform. At last, we carry out a user study, where four groups carry out a travel planning task using uTablePlatform. Results suggest that the UI management technique we present can help people complete individual or collaborative tasks effectively on tabletop.

Abstract:Activity recognition has many potential applications in pervasive computing field, such as medical care, intelligent home/office, business service, etc. Recently, sensor based activity recognition has attracted much attention, due to its advantages of ubiquity and less intrusion. This paper studies sensor based activity recognition. In particular, it recognizes high-level user activities from multi-mode sensor data using machine learning methods. This paper proposes an activity recognition framework, which makes use of low-level wireless network signal and acceleration data to answer questions like “where is the user”, “what is the user doing” and “what is the user going to do”. Three main fusion algorithms are included in this framework. Considering the trend of pervasive computing, this paper uses the mobile phones and their embedded sensors to collect activity information. Effectiveness of the proposed algorithms is confirmed on real world collected dataset.

Abstract:Existing visual analytic systems have deficiency in supporting the iterative process of analytic reasoning, which bring heavy memory burden to users. To solve these problems, this paper studies and designs a visual analytic system based on interaction history on-line tracking mechanism. Firstly this paper analyses the interactive task model, secondly studies the interaction history on-line tracking mechanism of the system. Thirdly the paper studies the action predicting method based on interaction history, and then put forward the visual metaphor self-adapting layout mechanism. Finally, it gives the application instance of the system and the experiment evaluation. The result of the evaluation experiment demonstrates that the system can provide natural and effective support for the process of analytic reasoning.

Abstract:To investigate effects of the frequency parameter on the curve shape, this paper analyzes the shape features of the cubic ω-Bézier curve by using the method based on the theory of envelope and topological mapping. Necessary and sufficient conditions are derived for this curve having one or two inflection points, a loop or a cusp, or be locally or globally convex. Those conditions are completely characterized by the vertex of the control polygon and frequency parameter. Furthermore, it discusses the influences of frequency parameter on the shape diagram and the ability for adjusting the shape of the curve.

Abstract:We explore for designing an algorithm to construct NURBS surfaces with approximate minimal area which interpolate given boundaries, and hence supply a gap that the current NURBS system is unable to design minimal NURBS surface effectively which is urgently required for engineering. Applying multifold techniques of NURBS surfaces such as knot-inserting, Hybrid polynomial approximations, etc., one can convert a NURBS surface into Bézier patches which are relatively simple to get minimal surfaces; then by using iterative method of optimizing the control points of each sub-surface and updating the whole surface continuously, one can obtain approximate minimal Bézier patches with high-precision successfully. In the end, one could choose the corresponding iterated approximation algorithms according to various requirements of different users, to compute the NURBS surfaces with approximate minimal area which satisfy boundary position constraints.

Abstract:In this paper, a light weight architecture modeling approach based on feature wires is proposed, which is called Fe-Wires. The approach is a procedural modeling method based on shape grammar. In the method, 3D models are reconstructed based on inputting and manipulating feature wires of architecture objects to overcome the interaction weakness of original rule-based approaches. To interactively tuning complex architecture models, we present a global deformation model which uses axis-aligned spring constraints. Different from previous nonlinear optimization methods, our approach only need to solve a convex quadratic problem which can be efficiently solved in real time. The presented experiments show that our method can be applied on the real applications for interactive modeling inner and outer part of buildings.

Abstract:A new viewing method is proposed to automatically produce good viewpoints, to implement the progressive investigation from global to local for a 3D model. With hierarchical skeletons, the method can efficiently manage the global features, local features and details, and guided by the skeletons, produce good viewpoints to view these features with the interrelations among them represented well. In this way, the mechanism for human beings to understand a 3D model is well fitted, and many views can be reduced for promoting the understanding efficiency, in comparison with interactive selection of views. By experiments, compared with the general way to recognize a model through interactive selection of views, the new method can reduce the watching time by more than half, and reduce the recognition errors significantly.

Abstract:The 3-D digital geometric model processing, especially 3-D data acquisition, feature extraction, spectrum analysis, model classification and retrieval technology have been attracted much attention in recent years. A complete and optimal generating algorithm for a class of orthogonal function system, called V₂-System, is proposed in this paper. It has some advantages to analyze spectrum using V₂-System, such as save storage space and raise algorithm efficiency, etc. Firstly, exact expression and generating process of V₂-System is given. Secondly, a method of constructing quadratic surface model is proposed. This method construct quadratic surfaces with different accuracy and numbers freely on basis of triangular faces, consequently, it provides many conveniences for tests. Thirdly, it is given that a Bézier surface can be expressed with finite terms exactly in V₂-System. It suggests a technology on data compression. Finally, according to frequency spectrums of surface model, V₂-Descriptors and V₂-Distances are calculated. Based on V₂-Distances, different quadratic surface models are classified. It lays the foundation of model retrieval.

Abstract:Currently main display algorithms in realistic image synthesis of Computer Graphics and Virtual Reality couldn’t show the scene perception change of human eyes when the environment luminance varies dramatically. To improve the default, this paper proposes and realizes a new enhanced display algorithm based on human eye photoreceptor cell response mechanism to simulate light and dark adaption processes. First, our algorithm bases the different physiological response mechanisms of cone and rod cells of human eyes to establish the interaction model between the photoreceptor cell and its neighboring cell hiring bilateral filtering algorithm. Second, we build the dynamic model of luminance of objects in the scene under various environmental luminance changing with time. Then, by combining the tone mapping model of high dynamic range image, the empirical model of experimental physiology, adjusting the physiological parameters under various environmental luminance, the inverse response transform and chromatic adaptation transform, we finally realistically simulate the dynamic visual perception process of human eye during light and dark adaptation in 3D scene.

Abstract:Among the implementations of multi-touch technologies, computer vision is a widely used way to build the multi-touch table. The touch information are obtained through the processing the imaging of infrared lights which comes from the reflection on touching fingers. The processing includes finger area extraction, tracking, and rectification. Because of disturbance of environment and unequally spread of the infrared lights on the surface, the image processing methods of previous multi-touch toolkits have poor performance on the detection and tracking of touch fingers, and these methods have not considered the influence of finger moving and camera distortion. In this paper, we propose a multi-touch table toolkit named as MTDriver. In MTDriver, we use local extreme area to extract the possible areas of touch fingers, then we track the motion of touch fingers by the tracking method which use the adjacent frames information. Experiments show that MTDriver has higher precision, better performance, and more robust compared with the other toolkits.

Abstract:This paper introduces an interactive robotics modeling and simulation system called iRMSS (Interactive Robotics Modeling and Simulation System). iRMSS is composed of two sub systems. One is the 3D robotics modeling sub-system that facilitates the creation of segment-joint model and physical model for robots. The other is the interactive robotics simulation sub-system based-on Microsoft Robotics Studio that can provide dynamics parameters feedback. So that iRMSS has the advantage of rapid robotics modeling and flexible parameter feedback over the MSRS. An experimental case is showed to verify the system feasibility.

Abstract:Most existing researches on task planning in joint operation simulations are not based on a complete formal theory, thus can only support the description of low-level interactions among entities, but can not support the planning of multi-level organizational groups. The description logic of tasks is a theory for describing organization structures and collaborative behavior plans. This paper proposes a method named TPDLT (task planning based on the description logic of tasks), which provides a formal description of the task planning in joint operation simulations based on the description logic of tasks, and gives a definition of the completeness of task plans and a decidable theorem of the completeness under joint strategies. At last, a deduction algorithm for task planning is given. As illustrated by a case study, with the method TPDLT, the task interaction semantics in virtual groups can be described succinctly and accurately, and powerful service for the accomplishment deduce of tasks is provided.

Abstract:This paper presents a method of sketch-based human body modeling, which allows users to rapidly construct human model by sketching human skeleton and contours separately. The are two main features of our method: firstly, a multi-layered cross-section based human template is used to construct the result model, and a distance-based descriptor is computed in the model retrieval process to find the template most consistent with user’s sketch; secondly, a parameter-driven method of 3D human body modeling is used, it matches the sketch contour features and template contour features to compute 3D deformation parameters, by which the result model is constructed. The experiment result proves that the proposed method can rapidly construct 3D human body model which consistent with user’s intent.

Abstract:Markerless human body model reconstruction has been a hot research topic for decades, and it has many applications in human computer interaction (HCI), medical rehabilitation, sport training, and virtual human synthesis for games, cartoons and digital films. This paper presents a markerless 3D human body skeleton model reconstruction method. It takes input from multi-view video, and computes the visual hull of human body through shape-from-silhouette. It then extracts the 3D skeleton by a new hybrid 2-subiteration thinning algorithm, and registers the skeleton to a 3D human body skeleton model by a heuristic method. The experiment results show that although the intermediate results of the shape-from-silhouettes process and 3D thinning process contain much noise, the model registration method, which bases on a set of simple prior rules , can still get accurate 3D human body skeleton model.

Abstract:Viewpoint selection is an important research field in computer graphics, which analyses a model and generates viewpoints. These viewpoints fit well with human preferences, and contain as much as possible salient model features. But until now, the evaluation of viewpoint selection results has not yet been quantized. This paper designs and implements a quantized viewpoint selection benchmark. At first, it collects human viewpoint selection results of 30 persons observing 45 models, and defines the standard-human-viewpoint of each model according to these results. Then, it analyzes the consistence and stability of human viewpoint selection. At last, it implements five (three kinds) representative viewpoint selection algorithms, analyze and compare their results with standardhuman- viewpoints. It also analyzes the performance differences between each kind of algorithm dealing various types of models. The experimental results show that human viewpoint selection has good consistence and stability, and differs on various types of models. All tested algorithms don’t have great performance difference, but mutual information based method and viewing plane feature measurement method are a little better than others. Each kind of algorithm performs best at different type of models.

Abstract:A novel approach for 3D motion capture data retrieval based on the Self-Organizing Map (SOM) and PCA index is proposed. Firstly, the Self-organizing map topological features are enhanced before learning. Characteristic will be mapped to a main surface after training every motion. Then, principal component analysis (PCA) is applied to deal with the surface characteristics. To improve the retrieval efficiency, an indexing scheme based on the principal eigenvector is constructed. Introduction of SOM avoid directing contact with the raw data, Follow-up work are only based on the main surface. It eliminate the position information influence of different frame length of the motion characteristics. Experimental results show that the algorithm is effective.

Abstract:In this paper, we present a novel and fast algorithm for 3D meshes segmentation based on the geometric information of boundary. Firstly, according to the topological relationship of original mesh faces, we construct the dual graph, and set the weight of vertex and edge on the basis of geometric information of mesh faces. Using the k-way multilevel cutting method, we segment the dual graph quickly to obtain the pre-segmentation and initial boundary for each patch. Then we define the boundary strength function and the feature boundary of each patch, to construct a deformation contour on the patch’s boundary. By minimizing the energy of the deformation contour, the initial boundary is driven to approach the feature boundary. Finally, the original mesh is segmented into several meaningful patches in accordance with the minimal rule. The experiment results suggest that our algorithm is efficient and effective, and is applicable to a variety of triangular mesh models with prominent shape features.

Abstract:Scatterplots matrix is still one of the most popular and widely used approaches to explore multi-dimensional datasets with the advantages of simplicity and clarity. However, this technique is suffering from some shortages. It will result in clutter when displaying large complex datasets, because the data points overlap each other. In addition, it’s difficult to convey more information except the distributions between two dimensions. This paper improves and extends the current scatterplots to address these shortcomings. a) It glances at the scatterplots matrix and emphasize its single unit by overview + detail. b) It uses clustering algorithm to divide all the points in a scatterplots into several groups to avoid confusion. c) Bar axis instead of line axis is used to illustrate the density on each dimension, conveying more information. d) Histogram is another approach to express the same data feature with bar axis. e) Several interaction techniques are adopted to adjust the visualization. Finally, some scenarios are created to argue that this approach is available and effective. This approach is helpful in visualizing and analyzing the large complex data sets in the area of finance and industry.

Abstract:It is very important to select the most suitable motion words from surrounding text to describe the persons’ motion expressed in images during semantic understanding. Traditional approaches often learn a generative model to denote the occurrence probability between visual objects & motion and their corresponding annotated tags, and the learned model is then utilized to recognize persons’ actions in a new image outside training samples. However, all of existing approaches neglect the grouping effect of high-dimensional heterogeneous features inherent in images. In fact, different kinds of heterogeneous features have different intrinsic discriminative power for image understanding. For instance, the features extracted from arms are most discriminative to human waving motion. The selection of groups of discriminative features for motion recognition is hence crucial. In this paper, we propose an approach to select discriminative subgroup visual features from high-dimensional pose features by Group LASSO during the learning of generative model in order to boost the motion recognition. Experiments show that the proposed approach in this paper can obtain better performance for the recognition of motions with large pose change.

Abstract:BLAS library is the most basic math library in high performance computing. Its performance has a great impact on the performance of supercomputers. With the multi-core technology development, BLAS’ multi-core parallel performance has become more important than single-core performance associated with architecture. The experiment takes X86 multi-core processors like Xeon, Opteron series as platform for example, which are popular in HPC. It fully tests GotoBLAS, Atlas, MKL and ACML BLAS libraries of all 1,2,3-level functions, and covers different scales and multi-core parallel aspect. BLAS source code, material and papers, test results are used to analyze the way of BASL optimization and parallelism, and which platform they are suitable for. Then we will provide some useful suggestions for the use of BLAS, BLAS optimization method or even the development of high-performance CPUs. It was found that compared with a logically powerful and complex CPU, a CPU which has larger and better caches, wider memory bandwidth, smaller memory latency, higher core frequency can often obtain better performance in HPC applications. At the same time, the condition of X86 platform is also a good example for other architectures.

Abstract:Multiple parallel transactions in new-type distributed software result in that the events produced by them are randomly ranked. If the tokens of these events are incomplete or unavailable, they can’t be distinguished to belong to which transaction. In this paper, the problem of stripping events with incomplete tokens is converted into maximum-weight perfect matching of bigraph system. If the transition time among these events is independently and identically distributed, all possible states (events) are separated into multiple cutsets, each of them becomes a bigraph system. The improved algorithm of maximum-weight perfect matching is adopted to finish respective matching, and then the matching results are spliced to gain the most possible footprint sequences produced by these transactions. Simulation experiments confirm that our method can effectively strip transaction footprints with incomplete tokens. Compared to traditional bigraph matching algorithm, the improved one has higher efficiency.

Abstract:Accurate prediction of the running time of parallel jobs under different computing resources is the foundation of many job scheduling approaches. A job performance prediction method based on the Performance Skeleton is proposed to avoid the inaccuracy of historical and modeling analysis prediction methods in heterogeneous clusters. To record the running trace, a method is designed to access all communication traces during the runtime. To merge these traces, this paper designs a trace-merge algorithm to structure the communication traces. To compress the circulatory traces, which is the most central and difficult, this paper converts it into a circular sub-string compressing problem, and proposes an algorithm based on the suffix array. Its performance is theoretically and practically better than the existing algorithms. To automatically reconstruct the Performance Skeleton, it solves the scalable problem of calculation and communication time. Experimental results show that these methods can accurately estimate the running time of computing jobs. The error is less than 3% for homogeneous clusters, and 10% for heterogeneous clusters.

Abstract:The general purpose GPU computing technology has been successfully used to accelerator many important applications. Though researches have designed many programming models and performance models, the amount of effort required to optimize the performance of applications on GPUs is still very high. In this paper, we propose a GPU computational model based on the ability that how much an algorithm could hide the latency. The experimental results show that our model could predict three matrix multiplication algorithms well.

Abstract:To study the properties of metal under extreme conditions, a parallel program for 3-dimensional dislocation dynamics (PDD3D) is developed based on JASMIN. By integrating effective distributed data structures, a scalable fast multi-pole method solver and a novel communication scheme based on ghost, PDD3D achieves good parallel scalability. In a simulation of a model containing 30 million dislocation segments, the parallel efficiency of PDD3D of 1024 CPUs relative to 128 CPUs is 81%.

Abstract:Difference stencils are fundamental computations throughout a broad range of scientific and engineering computer programs. In order to optimize data locality and communication overhead, this paper proposes a novel alternate tiling stencil algorithm on distributed memory machines by exploiting the property of the iterative algorithm. The serial execution process of this iterative method is given, which introduces the sequence of iterative space tile as the sequence of execution, and uses time skewing technique to divide iteration space. In this process, nodes of the tile can be traversed many times to improve data locality. The parallel algorithm based on iteration space tile technique is presented, which uses an improved polyhedral model to implement the iteration space tiling algorithm and reorders the tiles of iteration space to reduce cache misses, and the cost of communication and synchronization. The theoretical comparison is given between alternate tiling and other parallelization techniques. Finally numerical results are presented to confirm the effectiveness of serial and parallel execution models of alternate tiling finite difference stencil algorithm, specifically compared with domain-decomposition and red-black iterative methods, and show that the new parallel iterative method has a good data locality, parallel efficiency and scalability.

Abstract:With the development of supercomputers, the CPU core numbers of which come to several hundreds of thousands, and on which the complexity of the applications run are increasing. Therefore, in order to optimize the source code of the programs, developers of parallel applications need to measure the performance of parallel applications and make a useful analysis, so that they can improve the performance of the applications. However, due to a substantial increasing of the CPU core numbers, performance measurement will produce vast amounts of performance data, and then, how to deal with massive data is a very critical problem for parallel performance analysis. A new approach, named Iterative based Clustering Approach for Parallel Performance Analysis (ICAPPA), is proposed for parallel performance analysis in this paper. In this approach, clustering method of data mining technique, which is used to processing massive data, will be carried out iteratively for the result in some conditions after previous clustering, to find out the dominating functions and processes of the parallel performance. And Bayesian Information Criteria (BIC) is applied to evaluate the result of clustering method. By using BIC score, whether iterative clustering applied to the result is reliable or not can be decided. And at the end of this paper, the validity of that approach is verified by experimental analysis.

Abstract:Many-Core CPU preferring software-managed memory hierarchies than the Cache memory hierarchies owe to the area and power consumption. Software-Managed memory hierarchies need soft explicitly managed the data placement and data transfer. This paper proposes an easy data tiling method compilation techniques for these large memory objects such as large arrays base on the program loop characteristic and the scale of program data. This method is easy carry out in the compiler and has the equal efficient as the loop and data tiling method. The experimental results of several benchmarks show that this method can outperform the loop and data tiling method when this method may acquire additional data locality in the memory on chip.

Abstract:The emergence of location-aware mobile social networking applications (SNAs) has gained considerable prominence in ubiquitous computing. However, the weak designs of privacy protection result in more potential risks that users have to face. This paper presents a deep insight into the designs of privacy protection, especially from the perspective of location privacy. Three main potential risks and some necessary privacy-related functionality in current applications are analyzed and summarized. And with respect to the risks and weak protection methods, we propose the designs of privacy protection in LaMOC which is a location-aware mobile cooperation system developed by us. The designs include both the policy-based mechanisms and computational approaches to protect users’ location privacy. Policy-based mechanisms enable users to fully control their locations disclosure while computational approaches can protect users’ real locations from malicious attackers. The designs provide a reasonable level of privacy protection in location-aware mobile SNAs to some extent.

Abstract:As an improved model of fuzzy Petri net, adaptive Petri net (AFPN) has got the learning ability from neural network. But AFPN still depends on offline training data, while actual environment is so complex, vague and changeful that AFPN seems slightly inadequate. This paper proposes an approach based on fuzzy logic and feedback theory to improve AFPN. The approach introduces feedback mechanisms into AFPN to enhance the adaptive ability in dynamic environment. In addition, the approach embeds fuzzy logic theory into the representation of context information. Thus, the uncertain context information management is more conformable with person’s sense. The approach is also able to learn the parameters of membership function by using the back propagation algorithm of neural network. At the end of the paper, an experiment is designed to demonstrate that the approach is feasible and effective in fuzzy reasoning.

Abstract:For geographic routing in 2D ad hoc networks, greedy algorithm is efficient. The next hop node is selected according to the distance to the destination. However, greedy forwarding fails when a message reaches a local-minimum. Face routing is used to solve these problems. Unfortunately, these results cannot be applied to 3D networks directly. We propose an algorithm GSG (Greedy Surface routing Greedy) for geographic routing in 3D environments. We partition whole network with 3D Restricted Delaunay Triangulation. Triangles and isolated edges are defined as 3D components on Surface. By means of identifying intersecting triangles and edges, efficient routes are constructed on surfaces by bypassing local-minimums. Simulation results show that GSG achieves good routing performance and scalability.

Abstract:Based on the balanced energy consumption, this paper proposes a multi-gateway nodes data collected routing algorithm based on load balancing tree (TBLB algorithm), which combines factors with robustness, data forwarding and anti-interference etc. On condition that multi-gateway, TBLB algorithm forms a kind of load balancing tree with the gateway node for the root node combining node energy and degree, which coordinates load balance between the nodes through the load balancing tree and effectively reduces the node energy consumption. In addition, nodes make path choice and gateway switching according to the path performance evaluation factor W to further reduce the communication overhead of network nodes and to improve the gateway node bottleneck. Simulation results show that TBLB algorithm can effectively equilibrium network load. TBLB also can improve network energy consumption and packets of gateway.

Abstract:This paper investigates resource allocation in OFDMA cooperative networks for maximizing the network data rate and providing QoS guarantees of best effort services and real time services. It defines a network utility function of data rate, delay, and packet loss with power constraints, and the relay and subcarrier dedicated to achieve the optimal utility will be selected for cooperation transmission. With the proposed resource allocation scheme for power allocation on user-relay pairs or only the user over each subcarrier, QoS guarantees of different applications will be achieved. The objective of resource allocation is to strike a balance between maximizing the data rate utility and minimizing the delay and packet loss. By using the optimization theory and related methods, this paper solves the utility function and derives the resource allocation result. Simulation results show that the algorithm can converge. And they show the effectiveness of the algorithm by presenting the power allocation result and the throughput performance and delay performance of the proposed scheme.

Abstract:Web news articles play an important role in stock market. Sentiment classification of news articles can help the investors make investment decisions more efficiently. This paper implements an approach of Chinese new words detection by using N-gram model and applied the result for Chinese word segmentation and sentiment classification. Appraisal theory is introduced into sentiment analysis and Na?ve Bayes, K-nearest Neighbor and Support Vector Machine are used as classification algorithms. This method is used for a Chinese stock news data set. The best accuracy reaches 82.9% in all experiments. Additionally, it develops a prototype system to demonstrate this work.

Abstract:In order to improve the efficiency of event detection from on-line news stream, we propose a new method to accomplish detection task with window-adding, named entity recognition and suffix tree clustering. In our method, we make full use of informative elements extracted from news (such as date, place, person and so on) to help detection task, and accomplish the detection efficiently with news characteristics matching, which decreases text similarity computation greatly. Experimental results show that our method improves on-line event detection performance, without sacrificing detection precision.