Research Of Multiple Object Tracking Algorithm Based On Feature Matching And Kalman Prediction

As one of the hot technologies in research of computer vision, video target tracking is widely applied in science technology, aerospace, defense, medical and other health and various fields of national economy, which has a significant practical value and broad developing prospects. In real life, the environment of the moving target is a constantly changing, such as light changing, target deformation and target covering, so these factors will give a powerful challenge for moving target tracking. This thesis is mainly divided into the moving target detection and the moving objects tracking. Moving objects detection is that the image of moving objects is extracted from a series of images, which is the foundation for follow-up moving object tracking; moving target tracking is mainly to make sure the precise location of the moving objects in a series of images.In the aspect of moving target detection, firstly, the development of the moving target detection and the algorithm of common methods and the principle of moving target detection including the optical flow method, frame subtraction and background subtraction method, is introduced, and analyzes the advantages and disadvantages of several algorithms. And on the basis of the analysis for the background subtraction method, the thesis puts forward a background subtraction method based on the morphology. In the process of extraction of the background image, firstly the two adjacent frame images are made subtraction processing, then the subtraction image is binaryzated. In order to avoid the void phenomenon of the binaryzated image effecting on results, the binaryzated image is filled by the morphology methods, and pixels of the background area detected is updated to the background image in order to get the ideal background image. Then the image of the background subtraction can make sure the moving region from the image can be obtained. And based on the proposed detection algorithm target’s minimum bounding rectangle can be outputted and the result can be gotten. It is the foundation of the subsequent work.In the moving target tracking, in order to simplify the complexity of the moving state of multiple moving targets, firstly, the thesis will divide the moving state into five basic moving states, in order to facilitate subsequent processing. Secondly, four features are extracted from the target including the image area of the moving object, the target colors mean value, geometric center of rectangle box outside the target and the ratio of width and height of the target rectangle, and the four features creatively make a metric function. In the metric function, the features fusion coefficients of four features are adaptively adjusted with the different contributions of the features in the target matching. The target matching is made by the metric function and the threshold value defined. Thirdly, the principle of Kalman prediction is introduced, and this theory of Kalman prediction is applied to the moving target matching when the target is covered in order to make sure the position of the target, so that if the target is covered, the target tracking still can continue, at last, the simulation result of the moving targets tracking is given.