Pallet Detection Based on Halcon for Warehouse Robots

Abstract

Pallet detection is the key step of cargo handling for warehouse robots. In order to improve the recognition rate of pallet detection due to the influence of complex background, a pallet detection method based on point cloud is proposed. In this method, time-of-flight (ToF) camera is used to collect the point cloud. ResNet50 neural network model which is provided by Halcon software is used for deep learning, and deep learning is used to extract the region of interest of pallet contour. The extracted region of interest is processed to obtain the regions of the pallet pockets, and the minimum rectangles surrounding each of the pocket regions are solved to obtain the position coordinates of the pocket centers. The experimental results show that the precision of pallet detection can reach 94.5%. This method has high recognition rate in complex background, and has reference value for the design of pallet detection system of warehouse robots.