Precise Localization with Depth Camera in Industrial Intelligent Logistics

With the prosperity of intelligent logistic systems in warehouses, large firms with sufficient funds can further improve the efficiency of their warehouse management, thus improving corporate profitability. However, it is difficult for small firms to find an efficient way to implement intelligent logistics with relatively low costs. In this paper, a method of locating pallets in the warehouse efficiently with segmentation and minimum bounding box is proposed. In the off-line phase, a segmentation network is trained based on a hybrid data set containing real-world and synthetic data. During the on-line phase, the network obtained in off-line phase is used to infer the RGB-D data collected. Then, a segmented RGB-D image that contains only pallets is obtained, with which the point cloud can be generated. The minimum bounding box can be calculated subsequently. Precise position and rotation angle can then be obtained effortlessly. Experiments are carried out to evaluate the performance of the algorithm proposed. The results have illustrated a promising application of the proposed algorithm featuring high efficiency, satisfying accuracy, and low cost.