Crowded abnormal detection based on mixture of kernel dynamic texture
4
Citation
22
Reference
10
Related Paper
Citation Trend
Abstract:
A novel method for anomaly detection in crowded scenes is presented. In our method, a new feature which named Mixture of Kernel Dynamic Texture was used for video representation. The MKDT method jointly models the appearance and dynamics of the scene. Based on this method, the abnormal detection includes temporal detection and spatial detection. The model for normal crowd behavior is based on MKDTs and outliers under this model are labeled as anomalies detection. Temporal anomalies are the events with low probability under the MKDT models. While spatial detection based on discriminant saliency is used to get a spatial detection map. The proposed representation is shown to outperform various state of the art abnormal detection methods.Keywords:
Kernel (algebra)
Kernel density estimation
Representation
Feature (linguistics)
Foreground detection
Texture (cosmology)
Real-time segmentation of moving objects in video sequences is a fundamental step for surveillance systems. One of successful methods for complex background is to use a multi-color background model per pixel, like Gaussian mixture models(GMM). However, the common problem for this approach is that it suffers from high computation complexity and is unfeasible in the distributed real-time surveillance system. Furthermore, the GMM method generally can not solve the problems such as ghost, shadow and the situation of illumination changes. This paper proposed an effective scheme based on edge-characteristic and inter-frame difference. Experimental results show that the proposed algorithm can get exact moving object from complex background accurately like GMM, meanwhile dramatically reduce the operating time, which is 30% of the GMM. Furthermore, the approach can effectively eliminate the distributions of background and the changes of illumination.
Foreground detection
Cite
Citations (1)
Foreground object detection in video is a fundamental step for automated video surveillance system and many computer vision applications. Mostly moving foreground object is detected by background subtraction techniques. In dynamic background, Gaussian Mixture Model (GMM) performs better for object detection. In this work, a GMM based Basic Background Subtraction (BBS) model is used for background modeling. The connected component and blob labeling has been used to improve the model with a threshold. Morphological operators are used to improve the foreground information with a suitable structure element. The experimental study shows that the proposed work performs better in comparison to considered state-of-the-art methods in term of error.
Foreground detection
Cite
Citations (18)
For the purpose of precisely distinguishing the true moving target and the background in video surveillance, many strategies based on both background and foreground modeling have been proposed recent years. In this paper, we presented an improved moving object detection algorithm based on kernel density estimation which has two features. First, we construct a novel background and foreground model based on the basic nonparametric kernel density estimation and a joint domain-range foreground model. The foreground model applied here assures a more accurate detection result especially with dynamic backgrounds and building background with basic kernel density estimation helps to reduce the amount of computational cost which is usually a large number in many of the exists background-foreground models. Second, we present a strategy using edge detection to adaptively updating the background. By taking this method, our algorithm carrying out a quite exactly detecting result while immediately adjust to the changes in the background model, such like illumination change, objects from movement to static or conversely. Experimental results show that our proposal efficiently suppressed the inaccuracy caused by multiple reasons.
Kernel density estimation
Foreground detection
Kernel (algebra)
Density estimation
Cite
Citations (9)
The proposed work is targeted toward improving the Gaussian mixture model (GMM) for the background suppression-based moving object detection. The GMM has been widely used for moving object detection due to its high applicability. However, the GMM cannot properly model noisy or nonstationary backgrounds and fails to discriminate between the foreground and background modes. The extensions to GMM provide increased accuracy in expense of complex implementation and reduced applicability. In response, this work proposes two simple improvements: 1) a novel distance measure based on local support weights and histogram of gradients to provide distinct cluster values; and 2) use of background layer concept to properly segment the foreground. The method also uses variable number of clusters for generalization. The main advantages of the method are implicit use of pixel relationships through distance measure with least modification to the conventional GMM and effective background noise removal through the use of background layer concept with no postprocessing involved. The extensive experimentations on various types of video sequences are performed to validate the improvement in accuracy compared to the GMM and a number of state-of-the-art methods.
Foreground detection
Cite
Citations (32)
In conventional Gaussian Mixture Modeling (GMM), the risk that foreground models convert into background models increases with the accumulation of the foreground model's weight under a constant learning rate. That makes the conventional GMM unable to deal with slow moving objects. This paper proposes a motion detection algorithm which fuses foreground matching into the conventional GMM. The motion information contained in foreground models is obtained by checking in real-time the state of each pixel. Foreground matching enables the GMM to deal with indistinguishable moving objects and greatly improves its tolerance to slow moving objects. The quantitative evaluation and comparison show that the proposed algorithm outperforms the conventional GMM by detecting up to 23.3% of true positives with an acceptable cost of time consumption and the number of false detections.
Foreground detection
Cite
Citations (1)
Detecting anomalies in power consumption among distribution network customers is a critical aspect of maintaining the efficiency and reliability of distribution networks. Current research in anomaly detection primarily relies on user profile clustering, which involves subjective outlier detection-based criteria sensitive to input parameters, making it difficult to obtain objective and reasonable detection results. Therefore, in this work, an anomaly detection method for power consumption among distribution network customers based on autoencoder and kernel density estimation is proposed to objectively quantify anomaly levels. First, a processing method for load increment curves based on autoencoder and seasonal decomposition is proposed to detect abnormal patterns within these curves. Second, a quantification method for anomaly levels of power consumption based on kernel density estimation is proposed on the basis of the abnormal patterns detected, which enables the objective differentiation of anomaly levels among different customers and time periods. Finally, case studies on the actual load data from distribution networks are performed to verify the effectiveness of the proposed methods. The simulation results demonstrate that the proposed methods can effectively detect anomalies in power consumption and objectively quantify anomaly levels among different customers in distribution networks.
Autoencoder
Kernel density estimation
Anomaly (physics)
Kernel (algebra)
Cite
Citations (0)
Maintaining a reference background resides at the heart of any video surveillance system. Dynamic background presents impediment in the establishment of an accurate background model for video surveillance. Existing approaches utilize both statistical and non-statistical techniques for maintaining an approximation of the background. Statistical methods are computationally intensive but produce accurate results. Mixture of Gaussians is an efficient adaptive technique for background modeling. Different variants of the techniques are given in literature. In this research study, various novel approaches were proposed and employed for background representation through mixture of Gaussian models. Subtle improvement in foreground detection is reported in some specific cases.
Foreground detection
Representation
Cite
Citations (6)
Aiming at the peculiarity of pedestrian in the road traffic, an effective pedestrian detection method was proposed based on an improved Gaussian mixture model(GMM) in 3 aspects: parameter updating, background estimation and foreground segmentation. The possibility of misjudging the static foreground as the background was reduced using a parameter updating model based on the image segmentation. The time of the foreground merging into the background was controlled applying the adjustment scheme of foreground merging time.The foreground segmentation condition was optimized by introducing the concept of average weight. The test results showed that the improved algorithm is better than the traditional GMM. It is characterized by good robustness and adaptability, able to detect the slow-moving even static pedestrian.
Robustness
Pedestrian detection
Adaptability
Foreground detection
Cite
Citations (2)
In previous surveillance applications, algorithms for background modeling based on Gaussian mixture models (GMM) needed to specify two parameters: threshold T, which determines a proportion of the data that should be accounted for by the background, and a learning rate alpha specifying speed at which the distribution parameters change [Stauffer, CVPR 1999}. In the Basic Background Subtraction (BBS), foreground objects are found by subtracting a static foreground image. In the proposed algorithm, BBS is applied using background obtained from GMM. This way, threshold T is replaced by a foreground-background separation threshold S. The advantage is that S is less sensitive than T. To make the model respond faster to changes, recent observed value of the most dominant background component is used as a current value for a particular pixel, rather than the component mean value. Quantitative and qualitative results show the advantages of the proposed technique compared to GMM models.
Component (thermodynamics)
Foreground detection
Value (mathematics)
Cite
Citations (1)
We analyse the interplay of density estimation and outlier detection in density-based outlier detection. By clear and principled decoupling of both steps, we formulate a generalization of density-based outlier detection methods based on kernel density estimation. Embedded in a broader framework for outlier detection, the resulting method can be easily adapted to detect novel types of outliers: while common outlier detection methods are designed for detecting objects in sparse areas of the data set, our method can be modified to also detect unusual local concentrations or trends in the data set if desired. It allows for the integration of domain knowledge and specific requirements. We demonstrate the flexible applicability and scalability of the method on large real world data sets.
Kernel density estimation
Kernel (algebra)
Density estimation
Data set
Cite
Citations (143)