An Efficient Real Time Moving Object Detection with Storage Reduction

Visual surveillance systems start with motion detection. Detecting a moving object is always a greater challenge from a real time system. Tracking a moving object adds further the complexity. In this paper, we propose three significant methods. A Background Subtraction method (BS), Storage Reduction (SR), and Mobile Alert (MA).Our proposed BS modelling defines to identify the foreground objects. Our proposed ST module defines, to store only the object involved process instead of storing the entire video process. And MA module defines to send an alert message to the admin once the object is detected. Moreover, we are implementing Real time moving object detection and background learning into a single process of optimization, a much useful feature to provide security where it is necessary. The new methods are integrated into the real-time object tracking system. Normally a motion object An Efficient Real Time Moving Object Detection with Storage Reduction http://www.iaeme.com/IJCET/index.asp 52 [email protected] detection method attempts to locate connected regions that define or relate the moving objects within the scene; like frame-to-frame difference, background subtraction and motion analysis. Many different options are also available. We are trying to detect the object location in a frame. Based on its locality in different frames the speed and time taken to reach the target by moving object is calculated. Key word: Efficient Moving Object Detection, Video Surveillance System. Cite this Article: M. Ramamoorthy, Dr. U. Sabura Banu, J. S. Praveen, P. Deepalakshmi and M. SRI RAM. An Efficient Real Time Moving Object Detection with Storage Reduction. International Journal of Computer Engineering and Technology, 6(8), 2015, pp. 51-62. http://www.iaeme.com/IJCET/issues.asp?JTypeIJCET&VType=6&IType=8 I.INTRODUCTION Visual or Video surveillance is an active research topic based on its growing importance for security in public and private places. They have long been in use to monitor security sensitive areas. The goal of video/visual surveillance should accomplish the entire surveillance task as automatically as possible. The capability of being able to analyse object movements and track their activities from image sequences is crucial for any surveillance system. Video Surveillance is currently in its third generation [1]. The aim of the third generation system is to allow the online analyzed video data to be used in generation of alarm to alert the human operators and for the effectiveness of the offline inspection. To achieve that, 3GSS will provide smart systems that will be able to generate real-time alarms defined on complex events and handle distributed storage and content-based retrieval of video data. To make the video surveillance systems “smart” it requires fast, reliable and robust algorithms for moving object detection, classification, tracking and activity analysis. It could be effectively used in public and commercial security, private security, smart video data mining, law enforcement, military security and etc. Video surveillance or video analysis has Object detection, Object segmentation or classification, Object tracking and behaviour recognition as its key steps for automated system [12]. Moving object detection is always the first step of any computer vision application (automated visual surveillance, video indexing, video compression, and human machine interaction) apart from being useful for object tracking initialization for a visual surveillance system. Moving object detection should aim to extract the moving objects that are of surveillance interest with background which can be static or dynamic [19]. Many of the algorithms use either temporal or spatial information in the image sequence to perform moving object detection; the algorithms that have been already proposed can be categorized into any one of the three most popular approaches. They are background subtraction [2–4], frame differencing [5] and optical flow [6]. Optical flow is a more complex method and it does not work in real-time. Updating the background model continuously and determining the background and foreground classification threshold is a major challenge in background subtraction. Due to dynamic environmental conditions and common difficulties like illumination change, repetitive motion from clutter pushes the moving object detection from background into a great challenge. M. Ramamoorthy, Dr. U. Sabura Banu, J. S. Praveen, P. Deepalakshmi and M. SRI RAM http://www.iaeme.com/IJCET/index.asp 53 [email protected] Moving object detection is generally achieved by object detectors or background subtraction [7]. Object detector is a classifier which scans the image with a sliding window and labels it as object or background. Classifier is built by offline learning [8], [9] on individual datasets or by online learning with manually defined frame at the initial time of starting the video [10], [11]. Whereas background subtraction [11] compares images with a predefined background model and detects the differences as objects. The assumption is that there is no object when the background model was defined [13], [14]. It requires training models which limit the applicability of the above methods in automated video analysis. Motion based methods can be used to detect object [7], [14] which will avoid training models/phases, in motion based method, motion information to differentiate objects from its background. In reality, foreground motion's possibility of complication with non-grid shape changes and background being complex with illumination changes and varying textures which will not help motion segmentation. Normally, the motion based detection will classify pixels according to the patterns of the motion, which is called motion segmentation. Background estimation will also rely on static data. Hence, this methodology will not be able to handle more complex background or moving cameras [15], [16]. This approach can be classified as motion-based methods category. It would help to solve the challenges mentioned above by having a low rank matrix which will have a vectorized frame approximated, a pixel wise background modeling and subtraction where, moving objects can be detected as outliers in this low-rank representation. Object detection and background estimation are carried out without training sequences and the low rank matrix representation of the background makes it flexible to accommodate the global variations. DECOLOR uses real time video as batch [22], Here we have used real time video live. The proposed system architecture using DECOLOR algorithm is depicted in figure 3. We have reduced the webcam storage by storing only the image when an object is detected. 2. METHODOLOGIES AND RELATED WORK There are various methods existing for object detection which includes object detectors, image segmentation, background subtraction, etc., [7]. We aim to segment objects based on motion information which comprises a component of background modeling. 2.1. Motion Object Detection Object can be detected by using any of the standard methodologies like background subtract, temporal frame, optical flow, statistical methods etc. Background Subtraction: This is the most widely used process in object detection by separating out foreground objects from the background in a sequence of video frames. Fig.1 shows the background is relatively static because of its low computation cost. It is done in a pixel-by-pixel fashion.