Convolutional Neural Network-Based Sub-Pixel Line-Edged Angle Detection With Applications in Measurement

High precision measurement is becoming an imperative requirement in many applications. A novel sub-pixel line-edged angle detection method based on convolutional neural network is proposed in this paper. The line edges of targets are accurately estimated by their geometric slope angles with an edge point located on the line. Specifically, the pixel level line-edged images are first obtained by image preprocessing. Then, two separate convolutional neural networks are effectively constructed to boost their discriminative capabilities for the sub-pixel line-edged angle classification. The pixel level line-shaped edge images are used as input and the final network outputs are the specific sub-pixel level line-edged angles. Finally, the sub-pixel level diameter measurements are precisely performed with the estimated angles. Compared with existing methods, the proposed method can estimate the sub-pixel line-edged angle with 0.1 degree accuracy in end-to-end way, even for the noisy images. Simulation results for angle measurement and the real-world experiment for diameter measurement reveal the validity of the proposed method.