【摘要】 从天基安全系统的信息采集与目标识别的角度出发,使用神经网络的方法对目标物的三维重构进行研究。实现了从物体二维图像到三维立体的神经网络结构,该神经网络由特征编码、循环学习、解码三个部分组成,并对网络输出的体素概率模型进行Delaunay三角剖分和Loop细分,最终得到了目标物点集致密、细节良好的重构模型。该方法应用于天基安全智能打击系统,有效减少了对照片数量的要求、减轻运算压力,提高侦察安全性。更多还原

【Abstract】 In terms of information collection and target recognition of space-based security systems, the method of 3-dimensional reconstruction using neural network is studied. The neural network structure from 2-dimensional images to 3-dimensional object is realized. It consists of three parts: feature coding, loop learning and decoding. The probabilistic model is outputted and the prime methods of optimizing it are Delaunay Triangulation and Loop Subdivision. Finally, a reconstruction model with detailed dense details is obtained. In the space-based security system, this method can effectively reduce the number of photos needed and calculating pressure, and improve the security of reconnaissance.更多还原