Detection and Tracking of Arbitrary Objects in Video

Detection and Tracking of Arbitrary Objects in Video Kleanthis Constantinou Abstract— Detection and tracking of arbitrary objects in video is a technique which detect object and an object tracker follows that object even when the detectable part cannot be seen. The goal to detect an object in video or image is to determine whether there are any defined object in the video and return their locations, for example the object can be individual team members in a video showing sports, and it’s also been useful for the police in hot pursuit of vehicle by detecting the vehicle while moves. In this paper includes an analyses a methodology for detecting and tracking arbitrary objects in videos and documentaries. This work will explain how a moving object can allow deriving and maintaining a dynamic template of each moving objects. INTRODUCTION This paper will examine and analyze the paths followed for the implementation of a system that makes the detection and tracking of an arbitrary object possible. In addition the paper will point out the importance of embedding such a system in surveillance systems enhancing the need of those systems upon collecting cohesive temporal information though such an implementation. Section II will distinguish need for implementing such a system and how it can benefit its host. Section III will be stating the structure and the techniques used to properly manage the events of tracking and detection of an arbitrary object. Section IV will refer to the variety of problems disclosed in detection and tracking systems such as operation interference, while in addition it will state the required precautions that need to take place in order to prevent any operation interference and allow the system to run efficiently and effectively enhancing its accuracy. Section V will briefly explain the different types of surveillance systems and how they can be accessible. Lastly Section VI will display the steps followed in a moving detection system. In Video analysis the first step is the detection of moving objects and the areas which can be used are surveillance videos, tracking and monitoring people and traffic, therefore in this section we will be stating some examples on how the system works from a camera view and how effective the system can react. II. Reasons The reasons for providing an algorithm to make possible the detection of video objects is due to the need of acquiring data to be forced as an input to a computer based vision application. The application’s goal is to rebut tracking objects in the scene considering parameters in the background and the camera. Background based variables include the variation of light and objects that can change their status from moving to stopped and vice versa. The algorithm consists of two parts, the object detection which is light in terms of programming and a second part which is based on a more sophisticated structure that functions behalf of detecting objects in videos. The process of locating and tracking a moving object in video over time can be done by using a camera. Detection and tracking does not satisfy the purpose of extracting information’s but also to make implementation of systems such as traffic control, security and surveillance, medical imaging, human computer interaction, video communication and compression, augmented reality and video editing possible. Establishing correspondence of objects parts between consecutive frames of video it is the main goal of the tracking. The task of this application provides us with data that are used to enhance lower level processing like motion segmentations and data extraction such as activity analysis and behavior recognition which categorized as higher level processing. Methods and algorithms of detection and tracking The tracking and detection methods are categorized based on how an application can use them. Generally object tracking systems are adequate for outdoor surveillances videos where tracking parts of an object is necessary for several indoor surveillance systems. It is necessary to distinguish objects from each other in order to track and analyze their actions reliably. The main methods for object tracking include firstly the correspondence matching points and secondly to carry out explicit tracking by making use of position prediction or motion estimation. The techniques used for designing surveillance camera systems include the use of stationary cameras to allow the segmentation of each image into a set of regions representing the moving objects by using background differencing, and by using the method of k-Gaussian expand the video processing and allowing process of real stream videos with time varying background and without dedicated hardware. Figure 1: Tracking block diagram The diagram above shows the main blocks followed for object detection and tracking, where foreground and background are the basis for defining images. The information extraction in this scenario includes object attributes and features that could be used in applications and real time video applications. The Methods which classified as point detectors, background subtraction and segmentation is object detection. The information expected to be derived from the tracker is the trajectory of the path which has been followed from a moving object over time by locating its position in every individual video frame. The use of detection and tracking algorithms include implementation of techniques such as: data mining neural network artificial intelligence wireless sensor network biometrics. IV. Problems and Solutions Based on statements made in section II, background changes refers to light changing scenarios such as an outdoor scene, clouds covering the sun and for an indoor scenario such as turning off the lights. By considering those two factors there is problem for an object to be detected and tracked. So the approach cannot be based on frame difference where frame rate it is also depended on the object speed. From this perspective the attention must be laid on the moving object detection based on the background suppression where background model is computed and evolved frame by frame. Clarifying that statement object motion is defined by the difference between the current frame and the background model. Apart from that there must be a high response rate between the changing nature of background and reliable background model computation. Then a model must deal with erroneous ghost detection which includes objects in background that appear as moving in order to be able to compute the differenc e between those objects original position and the position that those objects where projected to after performing motion. Another puzzling fact that makes the algorithm more difficult and not approachable were the existence of shadows and moving objects while the associated shadows are sharing the same features of visual such as detectability and motion, so when the background is updated, the shadows and the moving objects are detected and grouped at the same time. The tasks that are affected by shadows its object classification and the assessment of moving object. This kind of problem mostly affects a system that controls the traffic which is evaluating the trajectories of vehicles. To eliminate such problems the approach of shadow detection needs to be defined and suppressed based on a color analysis HSV space. Another thing that interferes with the processes of tracking and detecting objects in video is the availability of video sensor, the zoom capabilities and videos streams acquired by moving platforms. In such situations the background differing techniques cannot be used because they rely on stabilization algorithm for canceling the motion of cameras, and because the stabilization and the detection are based on the background and cannot perform perfectly since it requires stabilization algorithms in order to affine the perspective model for motion compensation where the quality of compensation depends on the observed scene. To increase the accuracy of detecting a moving object we used a stabilization algorithm that locates regions of an image where this region detecting the normal component of the optical flow field. Surveillance Surveillance systems is been used for monitoring of the behavior, activities or other changing information and more often of people for influencing, managing, directing or protecting them. Such surveillance system serving government and law to enforcement to maintain social control, giving the privilege to prevent or eliminate threats because of the services suck monitoring and recognition which surveillance systems provide. Types where this kind of program and technologies are used: Computers: where responsible for the monitoring of data and traffic through internet, which is categorized in real time monitoring Computer surveillance is used monitoring all phones calls, emails, web traffic; instant messaging etc. Telephones: the official and unofficial tapping telephone lines, the program which is on use for monitoring it is on real time. By using speech to text software creates this kind of algorithm intercept audio and then processed by automated call analysis program where search for certain key words or phrases. Social network analysis: Creating social map network based on data were collected from Facebook, twitter from social sites and from phones call records. Biometrics: this kind of technology its for human analysis for their physical characteristics such fingerprinting, DNA and facial patterns. The technique used is called facial recognition and is based on person’s facial features to accurately identify them from video surveillance. Aerial: Aerial: is an airborne vehicle surveillance which is collecting visual imagery or video. Because this kind of system extraction is high resolution imagery of identification object of extremely long distance it require to use a surveillance hardware such as micro aerial vehicle Data mining and profiling: Data mining is mathematical algorithm method and statistical techniques to identify previously unnoticed relationships within the data. And the process of assembling information about a particular individual or group is called Data profiling which is use of generate profile.. Such application is use for economic and social transactions where the amount of data is large where application is working by following the electronic trail. Every transaction nowadays is electronic, resulting in an electronic trail like credit card, phone card, rented video etc. The most common type of Surveillance systems include utilization of cameras in order to survey a particular space. Surveillance videos up until now consisted of systems analogous to three differentiated generations, 1GSS, 2GSS, and 3GSS. The first generation was used for controlling a room using various cameras at different positions where the role controller was a person. The second generation involved the use of digital and analog subsystems where digital video was focusing on real time detection consequently giving the video human operators for filtering out spurious events. The third generation systems provide end-to-end digital systems followed by today’s video object detection systems. Examples From Video analysis Crossing line detection: The object is detected when a moving object crossing the â€Å"safety† line through the video processing. The safety line can be setup base on the background and the various security zones in arbitrary shapes within the cameras view. So when the object crosses the line the program will automatically activate the alarm and the object will be marked with an alert frame so that the system will mark its moving trace and will alert security personnel to pay attention to the object recognizing it as intruder. Figure 2: moving object crossing the safety line Appearing detection: when an object appears within the camera view alert detects and identifies it as a moving object, if the object behavior is according to the pre-defined alert condition the system will alarm and detect its moving tracks. This system will automatically detect any moving object like human vehicle in a designated area. Figure 3: Moving vehicle Guarding region Entry detection: By setting various security zones in arbitrary shape with in cameras view and through the intelligent video processing technique, automatically will detect moving objects such as human animals, vehicle etc. and if the object does not met the predefined rules when they entered to the security zone then alarm will alert and the object will be marked with an alert frame. Figure 4: Security zone in arbitrary shape Leaving detection: Can set alert areas or regions when an item is removed from its region and indicate its track using alarm frame when the object is removed from it position. Prevent prison break and kids who left the safe place from the kindergarten. Figure 5: Alert area or region CONCLUSION In this paper we analyzed the fact that a system for tracking and detection is necessary for computer vision application implementations such as video compression, video surveillance, vision based control, human computer interfaces, medical imaging, augmented reality etc. this kind of systems provide key tasks for monitoring and controlling applications by providing input data to video databases such content based indexing and retrieval. Reference point [1].http://ieeexplore.ieee.org/xpl/login.jsp?tp=arnumber=784651url=http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=784651 [2]. http://arxiv.org/abs/1210.3288 [3]. http://www.google.com/patents/US20130322689 [4]. http://www.slideshare.net/yuhuang/object-processing11 [5]. http://www.cs.cmu.edu/~wdn/myresearch.html [6]. http://jivp.eurasipjournals.com/content/2013/1/42 [7].http://www.reoll.com/index.php?option=com_contentview=articleid=5Itemid=8lang=en [8]. http://en.wikipedia.org/wiki/Video_tracking [9]. http://en.wikipedia.org/wiki/HSL_and_HSV