Automating Image Quality Detection for Parcel Identification Systems: A Computer Vision Approach

This paper presents an overview of the computer vision methodologies employed in an image quality detection project for parcel identification systems. The core objective of this project is to minimize erroneous chargebacks that occur when the system fails to capture accurate carton label information. Our developed defect detection framework replaces the labor-intensive and error-prone human validation process, thereby significantly reducing labor costs and improving the accuracy of defect identification in parcel identification images. The system uses a multi-sided vision tunnel to capture images of cartons from all angles. However, various issues such as dirty cameras, out-of-focus images, overly bright camera flashes, and partial images can impair image quality, leading to failures in automated parcel receiving. Each failure incurs additional costs for the logistics provider and may result in unwarranted chargebacks to partners, adversely affecting relationships. To address these challenges, we propose a computer vision-based approach to systematically identify and exclude poor-quality images from chargeback datasets, preventing erroneous chargebacks. This approach not only enhances the accuracy of automated receiving operations, but also supports downstream analyses to identify locations with frequent image quality issues. By partnering with these locations, preventive measures can be implemented to improve parcel image quality across the network. The outcomes of this project aim to streamline the automated receiving process, reduce operational errors, and foster better partner relationships by ensuring fair and accurate chargeback practices.