To improve the encoding capacity and decoding accuracy of encoded marker points in close-range photogrammetry, a method of cooperative encoding and positioning corresponding circular markers comprising positioning crosses, initial numbers, and encoded characters is proposed. Gaussian filtering is used to smoothly preprocess the collected images to eliminate noise. The adaptive local threshold method is employed to segment the target to obtain the character area and cross mark area. TensorFlow-MLP (Multilayer Perceptron) neural network is trained using the character sample library to classify and recognize characters. Finally, the cross mark area is filled and repaired. Sub-pixel positioning is achieved through the gray square weighted centroid method. This type of cooperative coding sign is uniquely identifiable in practical applications with high positioning accuracy and accurate and efficient decoding.