Moving Target Detection And Tracking In Image Sequences Acquired By A Mobile Camera

Moving target detection and tracking in image sequences acquired by a potentially mobile camera is one of very significant and difficult techniques in an intelligent surveillance system. Being a very hard research work, it catches great attention of the academes around the world, and great efforts and resources have been invested recently. The research work, which is not limited a specific problem, involves many subjects such as computer science, optical science, mathematics, perceive and control science. Because of the great value in theory and practice, various new theories and algorithms have emerged in endlessly. However, generally speaking, there are still many difficulties in practice. Specially, those problems deteriorate due to complex background. The objective of this research was to present new methods to enhance the efficiency and accuracy of moving target detection and tracking, and to resolve the difficulties caused by target distortion, camera motion and occlusion.The research of the dissertation involves global motion estimation and compensation, image stabilization, occlusion detection and tracking, trajectory prediction, moving target location, region extraction, moving target detection and tracking.The principal research work and novelties are listed as follows:1. An original definition of signal similar is presented in order to build the criterion of image similar, and a novel image partition method named improved-quad tree is employed to split each frame of video. Those processing translate the 2D image calculation of motion estimation into 1D signal computation to reduce the computational complexity of motion estimation.2. A novel global motion estimation method based on the projection signal similar is proposed. Firstly, the detail criterion of image similar, which is based on the projection signal similar, is described. Secondly, a multi-scale image partition method named improved-quad tree is proposed to improve the initiative matching efficiency. Finally, the detail motion estimation theories based on image similar and improved-quad tree are presented to reduce the heavy-laden computations and to finish the image stabilization.3. A new global motion compensation approach based on the scene 3D depth reconstruction is presented. Considering a moving camera, it is hard to distinguish...