Real-time 3D human objects rendering based on multiple camera details

Multimedia Tools and Applications - 3D model construction techniques using RGB-D information have been gaining a great attention of the researchers around the world in recent decades. The RGB-D...     

基于融合学习的无监督多维时间序列异常检测

随着多云时代的到来,云际智能运维能够提前检测处理云平台的故障,从而确保其高可用性. 由于云系统的复杂性,运维数据在数据局部性和数据全局性上呈现出多样的时间依赖和维度间依赖,这给多维时间序列异常检测带来很大的挑战. 然而,现有的多维时间序列异常检测方法大多是从正常时序数据中学习到特征表示并基于重构误差或预测误差检测异常,这些方法无法同时捕获多维时间序列在局部性和全局性上的信息依赖,从而导致异常检测效果差. 针对上述问题,提出了一种基于融合学习的无监督多维时间序列异常检测方法,同时对多维时间序列的数据局部特征和数据全局特征进行建模,得到更加丰富的时序重构信息,并基于重构误差检测异常. 具体地,通过在时域卷积网络中引入自注意力机制使得模型在构建局部关联性的同时更加关注数据全局特征,并在时域卷积模块和自注意力模块间加入信息共享机制实现信息融合,从而能够更好地对多维时序的正常模式进行重构. 在多个多维时间序列真实数据集上的实验结果表明,相较于之前的多维时间序列异常检测,提出的方法在F1分数上提升了高达0.0882.

     

多数据流的实时聚类算法

针对当前对多条数据流的聚类算法不能兼顾质量和效率的矛盾,提出了基于相关系数的多条数据流的聚类算法,实现固定长度的在线动态聚类。算法引入衰减系数提高聚类质量,以相关系数作为流数据间相似度的度量标准,将数据流划分若干个数据段,以各数据流的相关统计信息进行聚类,得到实时的聚类结构。实验结果表明,算法有较高的效率、聚类质量和稳定性。     

基于质点的可变形体自碰撞检测

自碰撞检测是可变形体模拟过程中最耗时的环节,提出一种使用图形硬件的快速算法。算法以质点而非三角形作为自碰撞检测的基本单元,用球体包围以质点为中心的局部区域,再用AABB包围该球体的运动轨迹并将数据组织成纹理送入GPU,通过两遍离屏渲染计算出碰撞对集合及每个碰撞对的碰撞发生时间,算法复杂度为O(n)。实验结果表明,使用该算法在大规模布料模拟中检测自碰撞,效率较高。     

Image saliency detection for multiple objects

Multimedia Tools and Applications - Traditional saliency detection methods are designed only for a single salient object and cannot detect multiple salient objects in the image. This paper proposes...     

Online multiple objects tracking with detection reliability prior constraint

Multi-object tracking (MOT) is one popular topic in computer vision. It remains a challenging problem in complex scenes, especially of objects with similar appearance. In this case, many existing data association strategies, which link detections among consecutive frames according appearance and motion cues, may fail to track due to unreliable detections or confused appearance and motion. To solve this problem, this paper proposed a novel online multi-object tracking method with detection reliability prior constraint. Our method integrates the trajectory estimation and detection-prediction association into a unified framework. The detection reliability prior constraint is built with the Hankel matrix from object motion model. When we build the Hankel matrix, we adaptively select a set of previous frames to predict object states and calculate the associated weights between detections and candidate objects. Data association in MOT then is estimated by maximum a posteriori (MAP) in a Bayesian framework, accompanied with both previous trajectory and the current detection reliability. Experimental results using synthetic dataset and four public challenging datasets demonstrate that, the proposed method has a good tracking performance compared with the state-of-the-art multi-object trackers.     

摄像机运动下基于背景匹配的运动目标检测

提出一种摄像机运动下的运动目标检测技术。首先获取相邻两帧图像上的对应特征点的坐标后应用到仿射变换模型得到全局运动参量来描述背景图像的运动变化。然后根据估计出的全局运动参量通过双线性内插法完成背景匹配。最后采用背景匹配后的两帧图像差进行目标检测。实验证明本算法在提高处理的精确度和时效性的同时还可以处理摄像机发生较大幅度旋转运动情况下的目标检测。     

Optimal feature points for tracking multiple moving objects in active camera model

Object tracking is an important task in computer vision that is essential for higher level vision applications such as surveillance systems, human-computer interaction, industrial control, smart compression of video, and robotics. Tracking, however, cannot be easily accomplished due to challenges such as real-time processing, occlusions, changes in intensity, abrupt motions, variety of objects, and mobile platforms. In this paper, we propose a new method to estimate and eliminate the camera motion in mobile platforms, and accordingly, we propose a set of optimal feature points for accurate tracking. Experimental results on different videos show that the proposed method estimates camera motion very well and eliminate its effect on tracking moving objects. And the use of optimal feature points results in a promising tracking. The proposed method in terms of accuracy and processing time has desirable results compared to the state-of-the-art methods.     

Real-time haptic interaction with RGBD video streams

Video interaction is a common way of communication in cyberspace. It can become more immersive by incorporating haptic modality. Using commonly available depth sensing controllers like Microsoft Kinect, information about the depth of a scene can be captured in real-time together with the video. In this paper, we present a method for real-time haptic interaction with videos containing depth data. Forces are computed based on the depth information. Spatial and temporal filtering of the depth stream is used to provide stability of force feedback delivered to the haptic device. Fast collision detection ensures the proposed approach to be used in real-time. We present an analysis of various factors that affect algorithm performance. The usefulness of the approach is illustrated by highlighting possible application scenarios.     

Reliable detection and separation of components for solid objects defined with scalar fields

The detection of the number of disjoint components is a well-known procedure for surface objects. However, this problem has not been solved for solid models defined with scalar fields in the so-called implicit form. In this paper, we present a technique which allows for detection of the number of disjoint components with a predefined tolerance for an object defined with a single scalar function. The core of the technique is a reliable continuation of the spatial enumeration based on the interval methods. We also present several methods for separation of components using set-theoretic operations for further handling these components individually in a solid modelling system dealing with objects defined with scalar fields.     

Real-time tracking of multiple objects in space-variant vision based on magnocellular visual pathway

In this paper, we propose a space-variant image representation model based on properties of magnocellular visual pathway, which perform motion analysis, in human retina. Then, we present an algorithm for the tracking of multiple objects in the proposed space-variant model. The proposed space-variant model has two effective image representations for object recognition and motion analysis, respectively. Each image representation is based on properties of two types of ganglion cell, which are the beginning of two basic visual pathways; one is parvocellular and the other is magnocellular. Through this model, we can get the efficient data reduction capability with no great loss of important information. And, the proposed multiple objects tracking method is restricted in space-variant image. Typically, an object-tracking algorithm consists of several processes such as detection, prediction, matching, and updating. In particular, the matching process plays an important role in multiple objects tracking. In traditional vision, the matching process is simple when the target objects are rigid. In space-variant vision, however, it is very complicated although the target is rigid, because there may be deformation of an object region in the space-variant coordinate system when the target moves to another position. Therefore, we propose a deformation formula in order to solve the matching problem in space-variant vision. By solving this problem, we can efficiently implement multiple objects tracking in space-variant vision.     

Real-time triangulation of point streams

Hand-held laser scanners are commonly used in industry for reverse engineering and quality measurements. In this process, it is difficult for the human operator to scan the target object completely and uniformly. Therefore, an interactive triangulation of the scanned points can assist the operator in this task. In this paper, we describe the technical and implementational details of our real-time triangulation approach for point streams, presented at the 17th International Meshing Roundtable. Our method computes a triangulation of the point stream generated by the laser scanner online, i.e., the data points are added to the triangulation as they are received from the scanner. Multiple scanned areas and areas with a higher point density result in a finer mesh and a higher accuracy. On the other hand, the vertex density adapts to the estimated surface curvature. To guide the operator, the resulting triangulation is rendered with a visualization of its uncertainty and the display of an optimal scanning direction.     

Real-time SVD-based detection of multiple combined faults in induction motors

Early detection of induction motor faults has been a main subject of investigation for many years. Several approaches have been proposed for identifying one or more faults treated in an isolated way. Multiple combined faults on induction motors represent a big challenge since the reliable diagnosis of a faulty condition under the presence of two or more simultaneous faults is really difficult. This work introduces a novel methodology that merges singular value decomposition, statistical analysis, and artificial neural networks for multiple combined fault identification. Obtained results demonstrate its high effectiveness on detecting faulty bearings, unbalance, broken rotor bars, and all their possible combinations. The developed field programmable gate array-based implementation offers a portable low-cost solution for online classification of the rotating machine condition in real time. Thanks to its generalized nature, the introduced approach can be extended for detecting multiple combined faults under different working conditions by a proper calibration.     

Recognizing objects with multiple configurations

Computer vision has been extensively adopted in many domains during the last three decades. One of the main goals of computer vision applications is to recognize objects. Generally, computers can successfully achieve object recognition by relying on a large quantity of data. In real world, some objects may own diverse configurations or/and be observed at various angles and positions, and the process of object recognition is denoted as recognizing objects in dynamic state. It is difficult to collect enough data to achieve the sorts of objects recognition. In order to resolve the problem, we propose a technique to achieve object recognition which is not only in static state where the objects do not own multiple configurations, but also in dynamic state. First, we apply an effective robust algorithm to obtain landmarks from objects in two dimensional images. With the algorithm, the number of landmarks from different objects can be appointed in advance. A set of landmarks as a point is projected into a pre-shape space and a shape space. Next, a method is proposed to create a surface among three basic data models in a pre-shape space. If basic data are too few to create a surface or a curve, a new basic data can be built from the basic data. Then, a series of new data models can be obtained from these basic data in a pre-shape space. Finally, object recognition can be achieved by using the new data models in shape space. We give some examples to show the algorithms are efficient not only for the objects with noises, but also for the ones with various configurations.     

Detection of moving foreground objects in videos with strong camera motion

In this paper, we propose a novel method for moving foreground object extraction in sequences taken by a wearable camera, with strong motion. We use camera motion compensated frame differencing, enhanced with a novel kernel-based estimation of the probability density function of background pixels. The probability density functions are used for filtering false foreground pixels on the motion compensated difference frame. The estimation is based on a limited number of measurements; therefore, we introduce a special, spatio-temporal sample point selection and an adaptive thresholding method to deal with this challenge. Foreground objects are built with the DBSCAN algorithm from detected foreground pixels.     

Reducing hardware with fuzzy multiple signature analysis

Fuzzy multiple signature analysis seeks the middle ground between the complex implementation and high overhead of conventional multiple signature analysis and the less-desirable error detection rates of single signature analysis. This scheme relaxes the restriction of one-to-one reference-signature correspondence required by the conventional multiple signature scheme, yet retains significant fault coverage. It also avoids the high fault-grading CPU times of the single signature scheme     

Experiences in developing a fair-exchange e-commerce protocol using common off-the-shelf components

Fair-exchange is an important property that must be ensured in all electronic commerce environments where the merchants and the customers are reluctant to trust each other. This property guarantees that none of the transacting parties suffer because of the fraudulent behavior of the other party in the transaction. In this paper, we describe our experiences in developing a new e-commerce protocol to address the problem of fair-exchange for digital products. The protocol is based on a novel cryptographic technique that we had proposed earlier. We show how to use common off-the-shelf software components to develop such a protocol.     

Real-time hardware acceleration of the trace transform

The trace transform is a novel algorithm that has been shown to be effective in a number of image recognition tasks. It is a generalisation of the Radon transform that has been widely used in image processing for decades. Its generality—allowing multiple functions to be used in the mapping—leads to an algorithm that can be tailored to specific applications. However, its computation complexity has been a barrier to its widespread adoption. By harnessing the heterogeneous resources on a modern FPGA, the algorithm is significantly accelerated. Here, a flexible system is developed that allows for a wide array of functionals to be computed without re-implementing the design. The system is fully scalable, such that the number and complexity of functionals does not affect the speed of the circuit. The heterogeneous resources of the FPGA platform are then used to develop a set of flexible functional blocks that can each implement a number of different mathematical functionals. The combined result of this design is a system that can compute the trace transform on a 256 × 256 pixel image at 26 fps, enabling real-time processing of captured video frames.

Constraints on general motions for camera calibration with one-dimensional objects

This paper focuses on two problems in camera calibration with one-dimensional (1D) objects: (a) to find out the general motion patterns well suited for solving the calibration problem, and (b) to improve the robustness and accuracy of the method. Firstly, a sufficient and necessary condition for the solvability of 1D calibration with general motions is proved. Then the special motion of tossing a 1D object is provided as an example to illustrate the correctness and feasibility of this condition. After that some practical issues on obtaining the solution are inspected. By avoiding singularities, the precision and robustness of the method are improved: the relative mean errors are reduced to less than 5% at the noise level of one pixel which surpasses the state-of-the-art methods of the same category.     

Accelerated hardware video object segmentation: From foreground detection to connected components labelling

This paper demonstrates the use of a single-chip FPGA for the segmentation of moving objects in a video sequence. The system maintains highly accurate background models, and integrates the detection of foreground pixels with the labelling of objects using a connected components algorithm. The background models are based on 24-bit RGB values and 8-bit gray scale intensity values. A multimodal background differencing algorithm is presented, using a single FPGA chip and four blocks of RAM. The real-time connected component labelling algorithm, also designed for FPGA implementation, run-length encodes the output of the background subtraction, and performs connected component analysis on this representation. The run-length encoding, together with other parts of the algorithm, is performed in parallel; sequential operations are minimized as the number of run-lengths are typically less than the number of pixels. The two algorithms are pipelined together for maximum efficiency.