Faster RCNN for Printing Nozzle Detection in Complex Scene

In the process of 3D printing, due to mechanical error and cumulative error, the printing nozzle is prone to position deviation. In this paper, a method for 3D printing nozzle detection based on image enhancement and faster RCNN is proposed. First, images grabbed by cameras from different positions are manually labeled. Second, according to the characteristics of nozzle, a variant of white balance algorithm is employed to enhance the images. Third, the images with regional labeling are put into the deep learning network to train, which subsequently constructs a faster RCNN detector. The experiment results show that the method proposed in this paper is robust to illumination fluctuation and also adapts to complex scenes. Compared with the faster RCNN, it can effectively improve the performance of the nozzle detection.