19-10-2016, 12:55 PM
DETECTION OF ANOMALOUS EVENT IN SURVEILLANCE VIDEOS USING KALMAN FILTER
1459852301-conf.docx (Size: 134.14 KB / Downloads: 4)
Abstract Video surveillance has gained importance in security, law enforcement and military applications and so is an important computer vision problem.The localized nature of event allows for detection of changes or anomalies in activities.For a wide range of scenarios both motion and appearance information are considered,so as to distinguish different kinds of anomalies. The proposed method introduces kalman filter it is a mathematical algorithm used to remove noise and extract particular portion from the video and also detect the motion of humans.The main objective of the proposed method is to detect the human efficient way.Histogram of Oriented Swarms(HOS), applied to capture the dynamics.HOS together with the kalman filter are combained to build a descriptor that effectively characterizes each scene. These appearance and motion features are extracted for the detection of anomalous event to ensure robustness to local noise, increase accuracy in the detection of local, nondominant anomalies, and achieve lower computational cost. Global System for Mobile Communication(GSM) is an open, digital cellular technology used for transmitting mobile voice and data services is used to send message, when an anomalous event occur.
Introduction
Human activity classification is an important yet difficult problem in computer vision, whose aim is to determine what people are doing in surveillance videos [1]. The widespread use of surveillance systems in banks, ATM, Military or ICU. The task of automatically detecting frames with anomalies or interesting events from long duration video sequences has concerned the research community in the last decade.Event,especially anomaly detection in Restricted area is very important. Example for security applications, where it is difficult even for trained personnel to reliably monitor scenes.
The analysis of motion and behavior in Restricted areas is a challenging task for traditional computer vision methods. The barrier which is difficult to overcome are occlusions, varying densities and complex stochastic nature of their motions[5]. One of the complicating factors is computational cost, so it has to be kept within reasonable limits[5].This means that “anomaly” pattern in one video sequence may often be the “normal” pattern on another video sequence.
In this framework we propose a novel method for anomaly detection and localization that includes both motion and appearance information. We introduce Kalman filter to capture appearance, Histograms of Oriented Swarms (HOS), to capture frame dynamics. Social Force Model (SFM) describes the behavior of the crowd as the result of interaction of individuals[2]. We compute the social force between moving particle to extract interaction forces. In a crowd scene, a change of interaction forces in time determines the ongoing behavior of the crowd[2]. We capture this by mapping the interaction forces to image frames .The resulting vector field is used to model the normal behavior.[2]
Kalman filter is introduced to characterize the crowded motion for anomaly detection. It efficiently capture the motions of crowded scenes in surveillance videos. This method can be efficiently applied even when the motion in the crowded scene is non-uniform in space and time, and ”anomalies “ appear locally in a changing context.
II.RELATED WORKS
Ramin Mehran,A.Oyama,M.Shah [1], Have proposed a social force model,to detect the abnormal behaviors in crowd scenes.The ability of this method to capture the dynamic of crowd behavior based on the interaction forces of individual without the need to track objects individually or perform segmentation.It indicates that the method is effective in detection and localization of abnormal behaviors in the crowd.
Limin Wang, Yu Qiao,Xiaoou Tang [2], proposed a latent hierarchical model(LHM)for classifying complex activities.LHM is a hierarchical model with deep structure,which decomposes activity into sub-activities in a coarse-to-fine manner.LHM is flexible and effective to deal with the duration variation and temporal displacement of each sub-activity.LHM is more suitable for activities with longer and more complex temporal structure and gains considerable recognition performance improvement.
Claudio piciarelli ,C.Micheloni,and G.L.Foresti [3], Have proposed a technique for anomalous event detection by means of trajectory analysis.The trajectories are sub sampled to a fixed-dimension vector representation and clustered with a single-class SVM.The presence of outliers in the training data has led to a novel technique for automatic detection of the outliers based on geometric considerations in the SVM feature space.
Weiming Hu ,X.Zhou,W.Li,W.Lou,X.Zhang and S.Maybank [4], proposed and effective framework for tracking object contours.Color-based contour evolution algorithm which applies the MRF theory to model the correlations between pixel values for posterior probability estimation.Adaptive shape-based contour evolution algorithm,which effectively fuses the global shape information and the local color information.Dynamic shape prior model effectively characterizes the temporal correlations between contour shapes in periodic motions.It obtains more accurate contours than the existing autoregressive model.PSO-based algorithm can be effectively with contour tracking for videos with abrupt motions,and it performs the particle filter-based algorithm.
Shandong Wu,E.Moore,Mubarak Shah [5], Have proposed a novel method for detecting and localizing anomalies in complicated crowd sequences using a Lagrangian particle dynamics approach.Effectively used for chaotic modeling of a scene.Chaotic dynamics of representative trajectories to be used for probabilistic anomaly detection and localization .
Nicola Conci ,H.Ullah [6], It proposed a method for crowd motion segmentation and anomaly detection using alpha expansion based on graph cuts.It demonstrated the reliability of the method shoeing the results obtained with four video segments for crowd motion segmentation and anomaly detection with different motion features.Results have demonstrated that the method is efficient and robust in case of moderately dense scenarios.
N.P.Cuntoor,B.Yegnanarayana,and R.Chellapa [7], Changes in motion properties of trajectories provide useful cues for modelling and recognizing human activitiesThe localized nature of events allows for detection of subtle changes or anomalies in activities.An event probability sequence is computed for every motion trajectory in the training set using HMM.It reflect the probability of an event occurring at every time instant.The usefulness of the proposed event representation is illustrated using activity recognition and anomaly detection.
Du Tran ,J.Yuan,and D.Forsyth [8], Have proposed a novel approach for detecting complex and dynamic events.The relaxation from video subvolume to spatio-temporal path makes the method more flexible and hence well addressed to complex events which could not be precisely covered by subvolume.The global optimal solution of the Max-Path algorithm improves the smoothness of the event, thus eliminates the false positives and alleviates.Max-Path’s lowest complexity makes it efficient to search for spatio-temporal paths in a large 5-D space of spatiotemporal, scale, and shape.It is flexible enough to be applied to a wide class of events.
Alexei Gritai,A.Basharat,M.Shah [9], Presented a framework for unsupervised learning of a scene model that captures object motion and size at every pixel location.The proposed framework provides a means of performing higher level analysis to augment the traditional surveillance pipeline.Experiments on real videos have proven the effectiveness of the proposed approach for local and global anomaly detection.This approach can easily benefit from online learning and can also be used for conventional applications like predicting object path and scene exit points.
Vittorio Murino, R.Raghavendra,A.DelBue,M.Cristani [10], Have proposed a new particle advection scheme to detect global abnormal activity from crowded video scenes using PSO and SFM.Demonstrated the capability of the method in capturing the crowd events.It does not require the tracking of individuals nor to perform a learning stage.This method is efficient, robust and highly performing in detecting the global abnormal activities on very different types of crowd scenes.
III. TECHNIQUES
3.1 PREPROCESSING :
Default videos is taken as a input and Gaussian mixture model is used for background subtraction. The detection of moving object uses a background subtraction algorithm based on Gaussian mixture model . Morphological operations are applied to the resulting foreground mask to eliminate noise. In this, the real time videos which are converted into number of frames.In preprocessing process noise is reduced and then it smoothens the video.
3.2 HUMAN DETECTION:
Human detection is a challenging classification problem which has many potential applications. In default videos we use kalman filter to detect the human. It will detect the human by foreground detection.Distinguishing foreground objects from stationary background is both a significant and difficult research problem.
3.2.1.KALMAN FILTER
Kalman filter is a mathematical algorithm used to remove noise and extract particular portion from the video and also detect the motion of humans. Almost all of the visual surveillance systems first step is detecting foreground objects.Short and long term dynamic scene changes such as repetitive motions (e.g. waiving tree leaves), light reflectance, shadows, camera noise and sudden illumination variations make reliable and fast object detection,first step is the background scene initialization.
We implemented adaptive Gaussian mixture model mainly, The background scene related parts of the system is isolated and its coupling with other modules is kept minimum to let the whole detection system to work flexibly with any one of the background models. Next step in the detection method is detecting the foreground pixels by using the background model and the current image from video, due to camera noise or environmental effects the detected foreground pixel map contains noise.In the final step of the detection process, a number of object features (like area, bounding box)are extracted from current image by using the foreground pixel map. We use a combination of a background model and low level image post-processing methods to create a foreground pixel map and extract object features at every video frame. Background models generally have two distinct stages in their process: initialization and update.
3.3 APPEARANCE EXTRACTION:
Histogram of Oriented Swarms, is used to capture frame dynamics.It is used to detect the activities of humans and the motions will be detected.
3.3.1 HOS DESCRIPTOR
In order to form the HOS descriptor, we examine the evolution of the agent’s positions,determined by prey motion patterns and the forces affecting the agents.We modify Newton’s second law of motion by inserting an elementary parameter γ that takes into account the previous velocity values. Then, the acceleration of each agent at position is given by the vector sum of all forces acting on it, considering the fact that an agent’s mass equals 1, along with the γ -weighted velocity of the previous time instant. When pixel flow undergoes a sudden change, it will be captured by the forces acting on it ,so the influence of its previous value will be mitigated. As a result of the forces, the swarm follows accelerated motion and the velocity of agent. Therefore, the positions of agents are continuously updated and their new values are given for each spatiotemporal location.
CONCLUSION
This project introduces an approach for an effective video surveillance in the current system, this overcomes the traditional surveying where human intervention is needed and has to watch keenly for keeping track of the entire system, for input videos kalman filter and Histogram of Oriented Swarms (HOS) are used to detect human. We have introduced a unique technique which is major advantage to the old system. We proposed that real time videos are taken as input. It has a feature in which it sends GSM alert if there is any variation in the captured pixel. Also we are intent to dedicate this project to many important surveillance areas so that Security should be more powerful