Optimize Visual Hull With Concave Based On Depth Map

Reconstruction by depth value obtained by depth device is the focus in the area of computer graphic and virtual reality in recent years.The depth image brings the new view for the concave object to reconstruct.Traditional method based on color image can only recover the boundary of the object.But it can't recover the concave part of one object.Although reconstruction based on depth image can recover the concave part,but the result has bad boundary which means that the edge information lost seriously.In order to complete visual hull computation of object with deep concave area,we propose a method that fuse the depth and color image based on space subdivision.The fuse method uses the depth information to recover the concave part and the color image to recover the whole border.It resolves the problem that the method based on color image can't recover the concave area.Also it can keep the object edge very well.The fuse method performs well in the part of speed and quality.The content of the research includes: 1.The fuse computation in depth images from multiply views.The depth image can reflect the distance information of one part object.Therefore,it can help to construct the real object border and then reverts the concave part.While,one depth image can just get only one part information of one object.So,we should carry out our research by multiply view depth image in order to revert concave area more exactly.It can revert precise concave part by the effective integration of multiply view depth information 2.The fuse computation of visual hull based on depth image and visual hull based on color image.We can use the character of depth image to reconstruct visual hull based on depth image.The hull has the disadvantages of rough edge and unclear outline,but it can reconstruct the concave area of one object.In the same,we can use the feature of color image to reconstruct visual hull based on color image.The hull has clear outline but has no ability to recover concave part.In order to recover a concave object that has clear outline,we propose a method that fuse depth image and color image to compute visual hull by drawing the advantages of the last two kinds of hull.It overcomes the disadvantage that the method based color image can't revert concave area and completes fast and exact reconstruction of the object with concave area.3.Accelerated parallel processing of fuse algorithm.In order to accelerate processing speed,we propose a parallel scheme for the proposed fuse method under the CUDA platform based GPU.We change the child mission with parallel feature to Kernel and then combine the Kernels to a stream.The parallel scheme guarantees the real-time of reconstruction.Finally,we implement the algorithm and test it in the virtual experiment system.By our analysis and compare,the results show that the method can revert the visual hull of an object with deep concave area real-time and accurately.