关键词: 局部轮廓形状, 管路结构, 管路测量, 形状描述子, 反向传播神经网络, 机器视觉

Abstract: To solve the problem of automatic tube structure recognition,a recognition method for complex tube structures based on local shape feature was proposed.The recognition method first obtained the different tube structures' shape.Then the obtained shape was converted to a signal by shape descriptor.Batches of signal samples was collected and applied to train a BP neural network to classify the different tube structures.The training experiments showed that the recognition method had accuracy in 97%.In a certain application,multi-cameras were applied to recognize the tube structures and a voting method was also applied to increase the accuracy,which could be applied to reconstruct and measure multi-branch tubes automatically.

Key words: local shape feature, tube structures, tube measurement, shape descriptor, back propagation, neural network, machine vision