The Study Of The Selected Objects Tracking Algorithm In Real-time Video

Real-time selected objects tracking is an active problem in computervision and has a wide variety of applications, such as military, medical,commercial etc.In addition, the accuracy and stability of selected objects trackingdepend on selected objects tracking algorithms to a great extent. Most of thecurrent approaches proposed are mainly feature-based or motion-based.These approaches are reliable to track objects, but they have a lot of data tobe processed and they are too complicated to apply in real time. In this paper,an improved tracking algorithm is researched in details by analyzing thecurrent algorithms of objects detecting and tracking.The Cam Shift algorithm is a kind of color histogram-based trackingalgorithm. In this method, the projection of the color histogram is obtainedby the main color of the selected objects. The object's position is computedaccording to the horizontal and vertical projections of projected image.Although this kind of algorithm carries out the real-time tracking of selectedobject, it depends on the color information of the object too much, once theobject of close color appears in the video, it's easily to lose the trackingobject. This text put forward a kind of improved algorithm which gets theCam Shift algorithm associated with the object centroid mach. We set aproper proportion between color mach and centroid mach, when there are noobjects of close color to selected object in real video is depend on centroidmach, when existence is set the color match for lord, the centroid mach forassist. The experiment result appears, the improved algorithm is free fromthe influence of the shape and scale variety, and also make a good result ofthe close color disturbing, It has a strong robust and simple operate, easy torealize the real-time tracking. The step as follows:The first step: labeled the tracking object by hand in the sequencepicture (use the red rectangle), get the position and size information ofselected object, and compute its color histogram.The second step: compute the color histogram of current frame, get theprojection of the color histogram and set threshold value to segment. Whilecarry on this operation, the color space of the picture should first convertfrom the RGB for HSV which of smeller relativity, then get its binary imageby adopt adaptive threshold method to segment. The course of extractingshape feature is: after medium filtering, edge detection and threshold, theedge of image is obtained. Actually, we set a single window to show thebinary image of selected object, its initialization is a headstand picture, sostill need to rotate it to show the exact shape of tracking object.The third step: calculate the centroid of the selected object. In this paper,we use the connected region labeling algorithm to carry out the calculationof the centroid.The fourth step: confirm the object's position according to thehorizontal and vertical projections of histogram as well as object centroidmatch, and self-adjust the size of tracking window.The track procedure of the whole algorithm is depicted as above. Thisthesis also introduces relevant knowledge and the realization of the key stepsand the resolution of specific issues. The structure of this thesis is asfollows:Chapter 1 introduces some kinds of current track algorithms, andcompares their advantages and disadvantages, and point out the importanceand improvement orientation of the improved algorithm.Chapter 2 simply describes the Cam Shift track algorithm and itsimprovement.Chapter 3 expounds the realization of the track algorithm. Fist, thisthesis describes the track procedure of the figure examination. Second, thisthesis describes a series work that have to be done after selecting the tracktarget, including transforming the color space and dealing with theprojection of the target template and extracting the outline and computingthe centroid of the target. Last, this thesis describes realization of the mainsteps and the resolution of specific issues which has been involved in thetrack and matching procedure of the whole algorithm.In Chapter 4, this thesis selects three targets with different colors andshapes to perform the improved algorithm. It shows the results of trackingthe automatic adjustments of the windows ,the shape change of the targets,absence from the video capture scope and the situation with interference ofshelters. Experimental results show that the improved algorithm canautomatically adjust the size of the window according to the tracking targets,and can adapt to the complicate movement of the targets. It also obtainsbetter result of the situation with interference of shelters.Chapter 5 is the conclusion and forecast. It concludes the work of thisthesis and analyses the shortcomings of improved algorithm which could bethe future research interests.