论文部分内容阅读
为了克服现有逆向工程装备价格昂贵且不能满足实时测量的缺点,提出了一种新型成本低、扫描速度快的三维激光自由曲面扫描系统。针对该系统的曲面重构问题,提出神经网络曲面重构方案,网络的输入选取所获得的点云数据的X、Y坐标,网络的输出则选取点云数据的Z坐标。比较了径向基神经网络(RBFNN)和多层前馈神经网络(MLPNN)两种典型神经网络曲面重构方案的优缺点。实验结果表明:RBFNN对于离散点云的曲面重构精度比MLPNN重构的精度高,但RBFNN较MLPNN所需的隐层神经元个数多。
In order to overcome the shortcomings of the existing reverse engineering equipment which are expensive and can not meet the real-time measurement, a new type of low-cost, high-speed 3D laser freeform surface scanning system is proposed. According to the surface reconstruction problem of the system, the neural network surface reconstruction scheme is proposed. The X and Y coordinates of the point cloud data obtained from the input of the network are selected. The output of the network is to select the Z coordinate of the point cloud data. The advantages and disadvantages of two typical neural network surface reconstruction schemes, radial basis neural network (RBFNN) and multilayer feedforward neural network (MLPNN), are compared. The experimental results show that RBFNN has a higher accuracy of surface reconstruction than discrete MLPNN, but RBFNN has more hidden neurons than MLPNN.