A High Accuracy, Light Changing Insensitive, Anti-Interference Workpiece Localization Method Based on Light Weight Neural Networks

A CRTS-III ballastless track plate has been widely used in the railway infrastructure. The anchor sealing using automated industrial robots with computer vision is an essential step of track plates production. This paper proposes a novel visual localization algorithm based on one-dimensional convolutional neural network models with high-precision and strong anti-interference ability used for anchor sealing robots. Our method not only has strong robustness to various noise interference during image generation and transmission and oil pollution on the surface of the workpiece, but has strong robustness to changes of workshop's light without pre-set thresholds or any adjustive parameters as well. The adaptability to over bright or over dark production environments can effectively improve the adaptability of the industrial robot vision system to the production environment. In addition, the neural network model used in our method is small enough, which does not require high equipment computing power, especially GPU, and can run at high speed on a pure CPU industrial control computer platform.