【摘要】 为了提高指针表自动读数方法的准确度和便捷性,提出了一种基于卷积神经网络的渐进式指针表自动读数方法。首先利用Faster-RCNN(Faster-Region Convolutional Neural Networks)模型检测当前视野下指针表目标的包围框位置,得到目标图像,简化读数前设备安装与标定过程;然后利用SIFT(Scale-Invariant Feature Transform)、RANSAC(Random Sample Consensus)算法和透视变换得到无倾斜、无旋转的图像,克服指针表倾斜或旋转对自动读数的影响;再利用八邻域法检测指针表表盘,大津算法(Otsu算法)提取分割阈值,区域生长法提取指针;最后利用卷积神经网络识别指针表的大刻度示值得到指针表的粗读数,利用角度法得到指针表的细读数,完成高精度自动读数。实验结果表明,文中所提出的方法具有较好的准确性与抗干扰能力,读数最大误差低于0.7%。更多还原

【Abstract】 A gradually automatic reading method for the pointer meter using convolutional neural network is proposed in this paper to enhance the accuracy and operability of the reading procedure of the pointer meter. Firstly, a Faster-RCNN(Faster-Region Convolutional Neural Networks)model is used to determine the pointer meter position and obtain the target image from the current view field and simplify the installation and calibration before the meter reading recognition. Secondly, SIFT(Scale-Invariant Feature Transform)algorithm, RANSAC(Random Sample Consensus)algorithm, and the perspective transformation method are employed to calibrate the tilt and rotation of the pointer meter. Thirdly, the eight-neighborhood method is applied to detect the dial, while Otsu and area growth algorithms are used to extract segmentation threshold and extract pointer, respectively. Finally, a convolutional neural network is adopted to identify the large scale of the pointer meter and obtain the coarse reading of the meter, and then the angle method is applied to obtain the final accurate reading of the meter. The experimental results show that the proposed method has good accuracy and robustness. The maximum error of the presented method is below 0.7%.更多还原