Recently moving target detection and tracking of knowledge of this one conducted a holistic review, and read a few review of papers, so this blog be to make a simple record, several important concepts are described and Analysis. Michael Kors Bedford & Astor He did not go to research in this area now that recent research results. Because in my opinion, the main idea of the algorithm is the same, each method has its adaptation scene. Axial caught enough! Prospects detect this one, I would rather recommend parametric methods, Gaussian mixture model and a codebook methods are proven, excellent performance in the actual project, but you have to make some improvements to the algorithm according to the scene you need. This article is only the first draft, the back will be perfect to add some links to papers, in the hope that a motion detection conduct a complete review. 1. Introduction for motion detection security surveillance video majority has the characteristics of a single-lens fixed scene. The basic method of moving object detection method is usually 2.1 frame difference of two adjacent frame images at very short time intervals (typically used for Michael Kors Designer video surveillance cameras can shoot 8/12/24 frames per second), so change is not when the background very intense and fast when you can take advantage of the difference between the two images before and after to determine the Christmas Angles iPhone 4 4S Cases target painting in motion, by setting the threshold value can be changed to adjust the sensitivity of detection methods. Advantages: operation is small, simple, low noise, insensitive to light. Disadvantages: too slow can not detect an object, the object will cause internal voids, the object itself is not complete. Objects too fast, it will Christmas Beauty iPhone 5 5S Cases form two prospect areas. 2.2 Optical Flow image on the screen-by-pixel vector assignment first, when moving object appears on the display screen vector moving target total occupied pixels and background pixels there are differences between the bound, if there is no moving target It appears the picture of the pixel vector change should be smooth. It is precisely because there is such a difference, to take advantage of the difference detect moving targets on the screen pixels. Advantages: Suitable Michael Kors Hot Sale for constantly changing background scenes (moving camera), does not require a pre-training videos. Disadvantages: poor noise immunity, computing capacity, light on the relatively large. 2.3 flowchart frame difference background subtraction method Michael Kors Hamilton can be seen in two of the basic framework of the flowchart is substantially the same, difference is that there is a background subtraction background model, this background model is used and all the test frame difference operation, as opposed to inter-difference method to spend one frame and the next frame image differential operation. Principle 3 CodeBook background model 3.1 CodeBook the set \\ (X = \\ {x_1, x_2, \\ cdots, x_N \\} \\) is a video sequence to a position contiguous pixel color values at different times. Where \\ (x_t \\) as \\ (t \\) time sampled RGB vector value, \\ (N \\) representative of the number of frames used to generate background model. Let \\ (Cbook = \\ {c_1, c_2, \\ cdots, c_L \\} \\) codebook summary for each of the pixels forming clusters, which \\ (c_i (i = 1, \\ cdots, L) \\) indicates code book one codeword (codeword), wherein each code word consists of \\ (v_t = (R_t, G_t, B_t) \\) and \\ (u_t = \\) components. Where \\ (1 \u003ct \u003cN \\), \\ (v_t \\) indicates the codeword in the RGB color space in three axes to value, \\ (u_t \\) indicates \\ (v_t \\) statistics, which \\ (I_t ^ {min}, I_t ^ {max} \\) indicates the minimum and maximum brightness value of the codeword, \\ (f \\) indicates the frequency of the code word appears, \\ (\\ lambda \\) indicates the maximum code word Cheap Michael Kors appears time gap, \\ (p \\) indicates the time code word first appeared, \\ (q \\) indicates time code word last appeared. Construction codebook for determining whether a pixel belongs to a codeword in the book of the conditions are: 1) compare the color space from the \\ (colordist (x_t, v_i) \\), where \\ (x_t = ( R_t, G_t, B_t) \\) is the current image of the input color value of a pixel, \\ (v_i \\) for clustering pixels codebook a code word, the definition colordist follows: \\ (colordist (x_t , v_i) = \\ sqrt {|| x_t || ^ 2-p ^ 2} \\) where \\ (p \\) as \\ (x_t \\) projected value of the cosine of the angle under. If the above calculation of the distance to less than a certain threshold \\ (T_1 \\), then the next step is determined. 2) Brightness pixels must be within a certain range (\\ (I_ {low}, I_ {hi} \\)) inside. If a codeword pixels and code book, will have to update the code word, the color values to recalculate the weighted average. If the current pixel does not meet the above Michael Kors Christmas Cases two cases, the re-building of a new codeword. According to the above method to establish the codebook for each pixel, under such circumstances, the prospect of a pixel will be generated as a codebook codeword (a clustering unit). For those maximum interval appears \\ (\\ lambda \\) is less than \\ (N / 2 Michael Kors New Arrivals \\) codeword deleted. Two 3.2 yards a book using improved 1) Do Michael Kors Accessories not just consider single single pixel, taking changes in their surrounding Christmas Beauty iPhone 4 4S Cases neighborhood 2) RGB space to calculate the distance by the cosine is not appropriate. Consider using YUV Michael Kors Handbags space. 3.3 brightness color space (V): the most sensitive human visual brightness changes, the size of the light energy. Hue (H): that is, we often say that the colors for color naming. Saturation (S): reflect the vividness of color. 3.4 Codebook model ranks YUV block sometimes shake the camera will exist, and some pixels will make changes as the environment on a background of instability. Image meshing, mesh 5 * 5 per share a codebook, so you can eliminate local jitter. Another method is to line codebook (1 * 5) and column codebook (5 * 1) are considered separately, the codebook a pixel row or column codebook determination of prospects, compared to before the point of interest. It can determine the column codebook row codebook threshold adjustment. 4. The method of moving target tracking moving targets represent 4.1 target tracking, the representation of the goal is the foundation of target tracking. The most common form is based on objective representation. It can have the following methods: 1) the method point representation. The target represented by a single point, that is, its center to represent. For relatively large target can choose to use a set of points to represent. This representation for the target relative to the entire image is relatively small. 2) the geometry of the notation. By defining the oval or matrix Christmas Angles iPhone 5 5S Cases box to indicate the target area. This method is only suitable only representation of rigid body motion less prone to deformation. 3) skeleton notation. After Michael Kors Clutches the target contour through axis transformation, you can make the skeleton model object. 4) outline notation. The outline of objects represents a boundary object. Secondly, the appearance of the more important features as well as target notation representation. A kind of the following: 1) the target probability density representation. 2) Template representation. 3) active appearance model representation. The target's shape and appearance while modeling. 4) The multiview model representation. More than 4.2 features color combinations feature: Lab space edge feature: Not illumination changes. Texture: LBP feature, 4.3 tracking method 1) is based on the tracking point 2) is based on statistical tracking 3) 4.4-based contour tracking method based on the optical flow tracking in general, optical flow is due to the scene to move the foreground object itself, movement of the camera, or jointly produced by the movement of both. The calculation method can be divided into three categories: 1) matching method based on region or feature-based; 2) a method based on frequency domain; and 3) a method based on gradient; simply, optical flow is a moving object in space observation imaging plane pixel motion of 'instantaneous speed.' Research on optical flow is temporal variation in Michael Kors Satchels the use of image sequences and pixel intensity data to determine the relevance of each pixel position of 'movement.' Objective To study the optical flow field is obtained from the picture to the approximate sequence can not be obtained directly stadium. Premise optical flow assumptions: 1) constant brightness between adjacent frames; 2) adjacent video frames in a row to take the frame time, or movement between adjacent frames object comparison 'small'; 3) consistent space sex; that is, the same sub-pixel images with the same motion. Optical flow method for target tracking principles: (1) a continuous sequence of video frames for processing; (2) for each video sequence, the use of certain target detection, detection of possible prospects objectives; (3) if a frame appeared foreground object, find the key representative of the feature points (which can be randomly generated, can also be used to do the corner feature points); (4) any two adjacent video frames later, the search A key feature of the previous frame appears best location points in the current frame, resulting in foreground object position coordinate in the current frame; and (5) so iteration, target tracking can be realized;moving target detection and tracking