Palletized cargo is essential in warehouses and logistics facilities. But damaged pallets and improperly packed products may cause serious problems. This includes loss of products, risk of accidents, and difficulties with automation equipment. It is crucial to identify these problems at an early stage.
Any minor faults such as cracks in the structure of the pallets or instability in loading can lead to bigger problems during the transportation process, particularly within automatic handling devices such as robots used in forklifts or conveyor belts. The current inspection system employed is quite manual and cannot be scaled easily.
Although computer vision algorithms have found significant use cases in manufacturing applications where they are used for quality control through object detection, the problem faced by most algorithms is that of generalization, since most algorithms are trained using a specific purpose and cannot work effectively in other situations. This constraint is particularly apparent in actual warehouses like those run by our partner, KION Group, which have significant variations in terms of the type of pallets, loading circumstances, and possible damage instances, thus motivating this capstone project.