Research Of 3D Reconstruction Key Technology Based On Monocular Video

As the widely use of the three dimensional reconstruction based on monocular video, it has also been a important direction which is received much concern during the past few years. This thesis studies the whole process of the three dimensional reconstruction presents a new algorithm based on mapping.Firstly, we study the YUV Model for motion capture. Luminance information in Y can detect black and white easily, but it does not deal with the shadow good enough as it only has luminance information. However, U and V can remove the shadow effectively, but it can not detect the foreground like shadow, so this thesis compositing their advantages and presents a method base on dual threshold to gain the better outline effect.Then, we chose a new method to represent the human body on this thesis. For the silhouette, we use radius bins combined with clustering to represent the border of the silhouette and use the histogram to represent the result .For the 3D human body, we choose the represent based on Euler angle to reduce the influence of the length of limbs. The method meet the need of the subsequent algorithm.Finally, this thesis studied the process of Relevance Vector Machine and use it to train the 3D data of human .During the training , we improve the RVM and get a good method of three dimensional reconstruction.As seen in the results of the experiment, this algorithm can resolve the key problem of three dimensional reconstruction, and get a good result.