光学学报, 2019, 39 (12): 1212005, 网络出版: 2019-12-06
基于BP神经网络的结构光光条中心提取 下载: 1291次
Center Extraction of Structured Light Stripe Based on Back Propagation Neural Network
测量 结构光光条 中心提取 BP神经网络 网络训练 误差分析 measurement structured light stripe center extraction BP neural network network training error analysis
摘要
为了精确、快速地提取结构光光条中心,提出了一种基于BP神经网络的中心提取方法。给出了使用BP神经网络实现光条中心提取的基本原理、训练网络所需的理想中心点的求取方法,以及网络权值的调整算法。研究了隐含层神经元个数m、隐含层层数h,以及训练样本对中心提取精度的影响,结果表明:当m=3,h=1,训练样本为带有噪声的随机光条时,神经网络能够得到更好的光条中心。由对比实验可以看出,所提方法相较于Steger方法和灰度重心法的中心提取精度更高,而且对1280 pixel×960 pixel光条图像中心提取的平均用时仅约为0.04 s,为Steger方法的0.27%。所提方法具有高精度、高效率等优势,能够满足复杂光条亚像素中心提取的要求。
Abstract
To accurately and rapidly extract the center of the structured-light stripe, we propose a center extraction method based on the back-propagation neural network (BPNN). The basic principle of stripe-center extraction using the BPNN, the method that calculates the ideal center points for network training, and the network-weight tuning algorithm are presented successively. Factors affecting the center extraction accuracy, such as the number of hidden layer neurons m, number of hidden layers h, and training samples are investigated. The center-extraction results show that the network can achieve a better stripe center when m=3 and h=1, and the training sample is a random stripe with noise. From the comparison analysis, it can be concluded that the proposed method can achieve higher center-extraction accuracy than both the Steger method and the gray gravity method. The average center-extraction time for a stripe image with the size of 1280 pixel× 960 pixel is 0.04 s, which is only 0.27% of the time required by the Steger method. This further demonstrates that the proposed method has the advantages of high precision and high efficiency. Therefore, it is adequate for sub-pixel center extraction of complex light stripes.
李玥华, 刘朋, 周京博, 任有志, 靳江艳. 基于BP神经网络的结构光光条中心提取[J]. 光学学报, 2019, 39(12): 1212005. Yuehua Li, Peng Liu, Jingbo Zhou, Youzhi Ren, Jiangyan Jin. Center Extraction of Structured Light Stripe Based on Back Propagation Neural Network[J]. Acta Optica Sinica, 2019, 39(12): 1212005.