Research On Abdomen Reconstruction For Multi-airbag Mannequin Based On Stereo Vision

The human abdomen shape reconstruction is a crucial technology in a wide range of fields, such as personalized garment design, medical care, man-machine engineering. This thesis conducts a further study on abdomen shape reconstruction and abdomen deformation reconstruction according to the principle of computer stereo vision, digital image correlation method and the theory of biological intelligence. Based on the full investigation and review to precious works related to the techniques of human body reconstruction in terms of stereo vision system, the main contributions carried out in this thesis lie in:(1) The basic principle and method of human body3D reconstruction by stereo vision are introduced, and then, a detailed exposition and summarization about the techniques and applications of body shape reconstruction at home and abroad by means of stereo vision are provided. Still, the critical techniques and related research works of3D reconstruction based on stereo vision are outlined. In addition, the theory of digital image correlation method and its application in image deformation matching are also expounded. All these offer the theoretical basis and the fundamental route for solving the problems of3D abdomen reconstruction for soft mannequin through stereo vision in the research.(2) Trying to explore a novel approach for the static3D abdomen reconstruction that is suitable for real photographing condition aiming at recovering abdomen shape. With the explicit markers are woven actively in the tight dress on the part of belly, the proposed combing the stereo vision theory, taking full account of the frequent interferences from both illumination variance and blur noise in the process of photographing, an innovative illumination-robust and anti-blur descriptor is designed for solving the binocular images matching problem in stereo vision system. Meanwhile, the strict geometric constraints are involved for eliminating the error mapping pairs, more exact matching pairs obtained at this stage. Subsequently, these exact pairs are taken as seeds to produce dense cloud data by means of global seed growing algorithm for achieving the recover of3D belly panorama for soft mannequin. The experimental results reveal that its precision can up to the highly similar measurement accuracy of3D scanner, which can well satisfy the requirement of fashion designing and facilitates the costume design industry.(3) A further exploration concerning the abdomen deformation modeling is performed based on the above proposed stereo vision platform and method for a static belly shape. A methodology of3D abdomen panorama reconstruction is put forward in the course of the belly deformation. A novel strategy of adaptive ellipse subset relying on the deformation degree as well as a new improved fractal dimension on the basis of the subset area is designed for the aim of the integer-pixel displacement search for the same feature point between the abdomen images before and after deformation in sequence. Then, a mutual learning adaptive particle swarm optimization algorithm is employed at sub-pixel registration resolution stage to locate the sub-pixel precisely. Supported by the combined coarse-fine ideology, the corresponding points between images of before and after deformation are established exactly for accomplishing the3D abdomen deformation reconstruction. Testing on the abdomen deformation reconstruction for soft mannequin, experimental results indicate that under the guarantee of its measurement accuracy without any loss, the time-consuming of the proposed scheme is significantly superior to that of the conventional method, particularly, at the large number of interest points.(4) In order to overcome the cockamamie steps of traditional matching method further, the stereo matching of feature points by B-T immune neural network is completed. The both similar mechanism between the left-right consistency constraints in image feature mapping and the mutual mechanism assistance and inhibition of B cells and T cells is abstracted. Moreover, the inherent geometry property is also introduced in T cell layer so as to ensure the uniqueness. For the more precise result, in each sub-grid of a circle template, a new combined feature vector is produced by means of involving entropy signifying the structure feature of image, as well as the gray information. Finally, neural network algorithm is utilized to obtain the global optimization effect for the all feature points. The intelligent3D belly recover for soft mannequin is achieved in virtue of the flexible bidirectional regulation of B-T immune network and the well organization property. Experimental results demonstrate that the proposed approach greatly outperforms the traditional matching algorithm on time load under the condition of the same precise, which provides an revolutionary new idea and effective revolutionary for stereo matching in the future.Finally a conclusion is made for the whole contents of this dissertation, together with the perspectives of this field for the next step.