Automatic Drug Box Recognition Based on Depth Camera

Automatic drug box recognition is crucial in the construction of automatic pharmacy. Complex patterns and variable sizes make it difficult to detect the drug box. Besides, a large number of highly similar drug boxes lead to a low recognition rate. This paper proposes an automatic drug box recognition method. This method utilizes the depth information and color information from the depth camera to detect drug boxes and calculate the size of the drug boxes, which improves the success rate of the detection. Then, the drug boxes are recognized by the ensemble learning of the lightweight neural network and the attention module is added to the ensemble learning to improve accuracy. A refinement scheme of size and region convolution neural network (SR-CNN) is proposed to further improve the accuracy of detection and recognition. Experiments show that the proposed method can effectively detect and recognize the drug boxes.