Detection and recognition of Chinese character coded marks based on convolutional neural network

In close-range photogrammetry, it is required that the adopted coded marks must have unique identification number and can be identified as well as located accurately in the image. In this paper, a kind of coded marks with Chinese character as encoding characteristic is designed, and a detection recognition method for the coded marks is proposed based on convolutional neural network. Firstly, a virtual camera method based on the camera imaging principle is used to automatically generate large amount of simulative images of the designed Chinese character coded marks, which are used as training samples. These samples are used to train the convolutional neural network that is used as the recognition network of Chinese character coded marks. The real captured Chinese character coded marks in the measurement images are detected with a series of cede mark sifting criteria, and the identification number is identified with the coded mark recognition network. Finally, the ceded mark centers are located with the center location algorithm. The experiment results show that the proposed recognition network has strong ability for recognizing the Chinese character coded marks, the recognition rate reaches 97.67%. The proposed method is less affected by noise, projection angle, image contrast and brightness changes, and possesses strong robustness. The proposed method can accurately segment the Chinese character coded marks and the center location algorithm can accurately locate the mark centers. © 2019, Science Press. All right reserved.