Development of a Single Camera Machine Vision System for Automatic 3D Size Detection

Abstract

Abstract The rapid growth of global e-commerce prompts the need of efficient warehouse handling and logistics, forcing manual operations to be replaced with automatic systems. An example of automated solutions is the smart packaging system for boxes which can simulate optimized box arrangements in order to save space. Even though three-dimensional bin packing problem has been studied widely to optimize box arrangement, only a few studies have been done to automatically detect box sizes. Box size detection is important to increase the effectiveness and efficiency of the packaging process as well as providing fast and accurate input for the box arrangement optimizer. Therefore, this paper presents the development of a machine vision system for automatic box size detection in 3D to support a smart packing simulator. The system uses a single camera and a platform covered with square grids prints. The box size detection algorithm was based on the localization of the platform area and it works by applying the bilateral filtering, binary thresholding and morphological image transform for the square grids feature extraction. To measure the box size, the length, width, and height of the boxes were detected by referencing it to the count of the square-grids. The volume of the boxes was further computed from the three-dimensional values obtained. The performance of the algorithm was then evaluated by calculating the error against the true value obtained from the manual measurements. The average accuracy for box size detection was 94.3% and analysis shows that the accuracy of the model was highly dependent on the size of the square grids on the platform. The result shows great potential of using a single camera system for automated 3D box size detection.