| The occlusion detection of virtual objects and real objects is a core and difficult problem in augmented reality,and also a key issue in natural human-computer interaction.Most of the existing methods are not be widespread due to the expensive equipment and complicated implementation process.However,the current image-based method can only obtain the rough positional relationship between virtual objects and real objects,the accurate occlusion detection method for virtual objects and real objects is relatively deficient,which limits the applicability and universality of augmented reality.To this end,this paper combines the image scene to research the virtual objects and real objects occlusion,and the research results obtained are as follows:(1)A ViBe foreground detection algorithm is improved.Firstly,initialize each pixel in the image,using the color similarity between the pixel and the pixels in the sample set to determine the positional relationship in the three-dimensional space of RGB and obtain the position of the spherical center,the threshold value is determined by the variance of the sample value and the position of the spherical center in three-dimensional space,and the threshold is determined by the threshold radius.The number of sample points determines whether the pixel is a foreground pixel.The pre-attraction point is updated in combination with the conservative update strategy and the pre-attraction counting method,and the detected binary image is obtained.Then,the noise reduction and hole filling of the binary image are performed to obtain a foreground image containing the shadow,and the foreground image is obtained by combining the threshold and the region segmentation to remove the shadow.(2)A method of feature point number optimization based on local image elimination and extension is presented.Firstly,finding a coincident area of the foreground object and the virtual object by pixel-by-pixel processing.Then,the feature points are detected and eliminated in the region to obtain effective feature points,expanding the pixel on the partial image for the region for the intersection of the feature points of the intersection region;Finally,the nearest neighbor ratio is used to match the feature points in the left and right images collected by the binocular camera,the depth of the corresponding feature points is obtained by the binocular ranging principle,and the depth of the partial image is estimated according to the least square method.(3)A layout and construction method of virtual and real object space based on hierarchical relationship is obtained.Firstly,the normalized transformation method of the virtual real object coordinate space is obtained by the graphic scaling transformation strategy,and then the farthest and the closest surface of the virtual object is calculated relative to the camera viewpoint;then,by overlapping each of the partial regions respectively The depth of the foreground image is compared.On the basis of hierarchical arrangement,the determination of the occlusion relationship between the virtual and real objects and the fusion of the virtual and real scenes are completed.(4)A method of natural interaction using virtual objects and real objects occlusion relationship is implemented.Firstly,the collision relationship between the virtual object and the foreground object is estimated by the occlusion change of different virtual-real coincident regions in the continuous frames.Then,traversing the pixel to obtain the minimum matrix region of the virtual real coincidence region,and calculating the coordinate value of the central pixel of the region.Using the position and depth changes of the central pixels between consecutive frames,the vector direction is obtained to determine the direction of force acting on the virtual object,and then the motion state of the virtual object is determined.The above results not only enrich the method of collision detection,but also provide a way of thinking for natural interaction in virtual-real fusion. |
