一种异步执行改进三维重构的并行方法

An improved method for three-dimensional reconstruction of parallel asynchronous execution

虚拟现实技术是一种模拟现实环境的交互式三维动态场景仿真,其功能需求因应用领域的不同而显差异,其共性技术难点集中在虚拟场景的高效构建和复杂环境的实时绘制,而目前三维重构方法具有较高的计算复杂度和较大的数据处理规模,影响了重建的质量和实时性.针对以上不足,文中提出一种异步执行改进三维重建的并行方法,其算法核心是将多目视频流重建过程分配到服务器端和终端异步执行;接着针对较耗时模块引入CUDA(compute unified device architecture)并行技术,使得重构效率明显提升;最后,通过实验验证了文中方法能有效提高计算效率,减少终端节点与服务器之间的数据传输量,并在一定条件下改善重构的及时性.在移动网络环境中,该方法充分利用终端硬件设备,为室内动态三维场景的实时呈现提供了一种实现方式.

Virtual reality technology is a realistic simulation of interactive 3 D dynamic scene. Because of the different application areas,its functional requirements are different. Their common technical difficulties are realtime rendering of virtual scenes and efficiently building the complex environment. At present,the high computational complexity and large scale data processing influence the quality of three-dimensional reconstruction and real-time. To solve the above problems,this paper proposes an improved method for three-dimensional reconstruction of parallel asynchronous execution. Its core algorithm is used to assign the multi-cast video stream reconstruction processes to the server and the mobile terminal and then execute them asynchronously. And with the introduction of CUDA parallel technology for the more time-consuming modules and the system makes the reconstruction efficiency improvement more obvious. Finally,the experimental results show that the proposed algorithm improves the three-dimensional reconstruction,reducing the amount of data transferred between the terminal and the server node,and ensuring the reconstruction timeliness. In the mobile network environment,the method uses the terminal hardware efficiently. And it can be seen that the method provides a real-time rendering method for indoor dynamic 3 D scene.