摘要
针对传统基于辐射度算法的室内场景三维虚拟现实方法存在耗时高、建模效果差的弊端,研究基于三维视觉的室内设计虚拟现实方法,采用主动式、全方位立体视觉传感器采集室内场景的三维点云数据,基于点云数据进行室内场景内的物体几何关系以及摆放位置分布,完成室内三维场景的自主合成,对物体摆放位置分布实施训练,通过三维场景点云数据集获取相同类型支撑物中物体产生的位置,对数据实施归一化操作,采用高斯混合模型拟合这些数据,训练出三维室内场景中物件在支撑面中的位置分布模型。采用基于深度信息场景重构方法实现室内三维场景的虚拟实现。实验结果说明,所提方法重构的室内场景直观、视觉效果好,并且具有较高的重构效率和精度。
Since the traditional indoor scene 3D virtual reality method based on radiancy algorithm has the disadvantages of high time consumption and poor modeling effect,the virtual reality method of indoor design based on 3D vision is studied. The active stereo omni-direction vision sensor(ASODVS) is used to collect the three-dimensional point cloud data of indoor scene.On the basis of point cloud data,the geometrical relationship of objects in indoor scene is analyzed,and the locating place distribution of the objects is trained to realize the autonomous synthesis of indoor 3D scene. According to the point cloud dataset of3 D scene,the position of the object in the same-type upholder is acquired to perform the normalization operation for the data.The Gaussian mixture model is adopted to fit the data,and obtain the trained position distribution model in support plane of the objects in 3D indoor scene. The depth information scene reconstruction method is used to implement the virtual realization of indoor 3D scene. The experimental results show that the method has intuitive reconstructed indoor scenes and perfect vision effect,and has high reconstruction efficiency and precision.
出处
《现代电子技术》
北大核心
2018年第5期78-82,88,共6页
Modern Electronics Technique
