Studies On Motion Segmentation Using GMMLS Method

The development in image processing is growing fast and becoming very effective in human daily life. The domain of the image segmentation is one of the most challenging problems in computer vision. In our present work a study in the domain of motion segmentation using the GMMLS (Gaussian Mixture Model Level Set) image segmentation method is presented.In the earlier chapters of this thesis, a brief introduction on the history of image processing and its applications, especially in the field of motion segmentation is given. After, the GMMLS (Gaussian Mixture Model Level Set) image segmentation method is introduced. We preferred this segmentation method to others due to its unifying expression by taking into consideration all the factors that influence the image data distribution, i.e., the essential attribute of the independent region, the external lightning, imaging and combined observation; and the independent region data mutual interference.The GMMLS method has already been verified to be reliable in the case of the segmentation of a static image [1], and now the present work aims at applying it in the domain of motion tracking. At last, we compared the application of the GMMLS in motion tracking to the other existing methods.The Gaussian Mixture method was used as a method of classifying together the data with same characteristics supposed to belong to the same object, and then the maximization for an optimal approximation was done by using the Expectation maximization algorithm in the process of feature extraction. And the level set method was used as the segmentation method. The clear and detailed explanation about the method is fully given in the dissertation.The video used in the applications were recorded in different locations (indoor, outdoor, on grass and on water) with different moving objects (human and animal); the objective was to analyze the impact of these different conditions on the program execution time, rate and reliability. The analysis of the data that was statistically performed by using the statistical software SPSS reveres that this program is reliable and has high execution in the motion tracking (mean speed of 1.05±0.30frame/second). In addition, compared to the method used in tracking single object and determining its velocity proposed by Lokesh Peddireddi [126], our method was proved to be more reliable.