Video Of Moving Objects To Find, Track And Block Matching Algorithm

Moving object detection and tracking is a major issue in many vision systems. At present, Compress video has been applied to science study, industry and agriculture manufacture, military affairs, public security, medical treatment, education domain etc. The methods of the traditional object tracking are processed in non-compressed video. The different methods can been classified to four groups: region-based, contour- and mesh-based, model-based, and feature-based. They detect and track the moving object according to the features of the color and shape of the moving object. Since MPEG-1 (Moving Picture Engineering Groups) and MPEG-2 standard were published in 1993 and 1994. It has been studied that moving object is tracked under MPEG standard. Some methods were selected that the compressed data fluid of I frame and P frame are decompressed to become some general images, and moving object is detected and tracked under decompressed pictures. In order to find the shape, color and edge of the object, the method of detection and tracking moving object must complete self-train so that obtains a sample of the tracked object. Due to the shape of the object may be changed along with the motion of the object. We must complete a recognition procession by the samples in all kinds of possible instance. Accordingly, as the object is located in an inconsistent intensity of the light, the color of object will change along with the intensity of light. These problems bring much difficult for tracking method.A technique was presented that a frame subtracts its neighboring frame, their residual image shows that the background of the image is expunged and the edge of object is reserved. This principle is based on the playing speed of the video over 25/s, the background is static, the displacement of moving object is very little. The shortcoming of this method can only process the background of static situation and there are not the interference of other object and the noise. This method has been applied on the security of traffic. For example, the camera is posited at the crossroad and is fixed. Here the scene and road are not changed. Moving objects are only the voitures and passengeres. When there are not the voiture and passenger, the screen image is considered as a reference frame. After this procession has been done, the background is easily filtered, the edge of moving objects are reserved. There are some things with invariableness color, and only similar color objects in the background. We set a filter template. Others color pixeles are filtered and the image is smoothed for eliminating noise, then the remarkable color region of the image is considered as moving object. This method is applied to the recognition of the face. Off course, because maybe have yellow race, white or black, every color of skin must do self-train of the sample. The range of every color is obtained under all kinds of the light intensities. Here the face can be detected by the color filtering in the image.The motion of the object is relative to the background. It is difficult to detect moving object owing to the motion of the background. But there is a new method for detecting moving object, i.e. it computes the velocity of the background. According to the velocity of the background of the opposite direction, it can be fixed to a static background. Moving object can be detected. The disadvantage of the method is that the computation is very complex when the background is on a non-uniformity velocity.The object is composed of several color blocks. Their position can form an encoded queue. The moving object is detected by an algebra encode recognition. For example, the object is composed of four blocks, i.e. block A, B, C, D. Block A locates the left of block B and above block C. According to order, its code is ABDC etc. This code does not change its order after the object rotates, zoom, downsizes. The disadvantage of this method is that it will lose the target when the color is changed.A video fluid is composed of compressed data which include I(intra), P(prediction) and B(bi-prediction) frame. The data of each block of P frame is predicted by motion prediction and compensation from the data and motion vector of I frame. The data of each block of B frame is predicted by motion prediction and compensation from the data and motion vector of I and P frame or previous decompressed P and future decompressed P frame. The data of each block in P and B frame is obtained according to optimal block match technique. This procession produces correlation data between I frame and P frame, or between B frame and I, P frame. Particular, motion vector of each block is obtained from I or P frame.Since features of MPEG, our paper presents a novel method that directly detects and tracks the moving object in video from MPEG coded data fluid. It utilizes the inherence data correlation and the features of the compress video fluid. The detection is stated by region growing based on motion-vector homogeneous method. Motion vectors of moving object are extracted by a designed statistics filter. This filter mainly processes motion vectors of the macroblockes of moving objects' boundary. Moving object is tracked according to its global vector. The trajectory of object is corrected by DC and partial AC coefficient of DCT. Our method can fully make use of the temporal and spatial correlation between I and P frame, and effectively avoid the difficult of change of shape, color and background of moving object. Whole procession only needs simple calculation. Experimental results demonstrate the effectiveness of this method.Precise visibility measuring of billboard advertising is a key element for the organizers and broadcasters to make cost effective their sport live relay. However, this activity currently is very manpower and time consuming as it is manually processed for the moment. In this paper we describe a technique for detection of commercial advertisement in sport TV. Based on some a priori knowledge of sport field and commercial advertisement, our technique makes use of fast Hough transform and text's geometry features in order to extract advertisement from sport TV images. The experiments show that our technique achieves more than 90% accuracy rate.In tennis match, it is a very important event that the scoreboard appears and the score is changed, score content change and advertisement shows describes one approach for extracting and recognizing of scoreboard in sport match. We propose one technique that can inspect appearance of the scoreboard and advertisement block and change of score content. Two methods are selected for defining the sample of scoreboard and advertisement region. The first method defines sample database by artificial operation, the second method builds one sample database and one potential sample database by the sample's self- training method. Some potential geometric regions are obtained by using a simple fast Hough arithmetic. Through block matching of binary image the potential rectangle regions are recognized and classified to several classes of scoreboard and advertisement.Digital Speckle Correlation Method (DSCM) has been applied to an optical measurement, this correlation method is a block match technique of the digital image. Because the definition of the correlation function approximates Gauss distribution and it has a character of the single peak. At present, there are the entirety searching method, the cross searching method, the mountain climbing method for searching for the pixel displacement etc. According to character of the single peak of the correlation function, we select the interpolation algorithm. The data of the interpolation are from the neighborhood matrix of the maximum correlation coefficient. The benefit of the interpolation is low computational quantity and anti-noise. The gradient method processes a uniform region of original image and changed image. It found on a displacement field based on the difference and distortion of two regions. The character of this method is that the consuming time of computation is a moderate and it is very good for micro-transmutation. [83] introduces a gradient algorithm based on subpixel displacement of DSCM considering the statistical characterization of micro-region. It defines a statistical function and takes the point of maximum or minimum as subpixel displacement. We ameliorate the cross searching and the mountain climbing method. This paper discusses the theory precision of the quadratic interpolation and the gradient method from mathematics theory. And we ameliorate their formula from discrete integer aspect. The experiments show that the computational results of new formula are more stable than general method. The variance of computational result of new method decreases 0.5%～20%.The experiments show that the computational results of the new formula are more stability than the general method. These methods will be able to effectively apply to the industry measurement.