Design and Implementation of a Parcel Sorter Using Deep Learning

Automation in industrial environment reduces the cost of the operation while increasing the overall performance. Having an automation mechanism in the e-commerce warehouses to sort the parcels based on their destinations or shipping method will reduce the parcel processing time significantly. To automate parcel processing in Digikala's warehouse, a parcel sorter system is designed and implemented. In this system shipment method of the parcel is indicated by a set of markers. A computer vision system is developed to identify these markers using deep learning algorithms. The parcels are identified while they are moving on the conveyor belt in a relatively high speed (1 m/s). The computer vision system is capable of processing 1.3MP pictures in real-time with a rate of 100FPS. To sort the parcels an omni wheel roller mechanism is designed and utilized. To achieve the best results in a practical environment, a gap optimization mechanism and pack positioning conveyor are implemented and placed before the sorter. This system is successfully installed in the Digikala's warehouse.