Rapid e-commerce growth has increased the need for accurate parcel dimensioning in last-mile logistics. Package size is often recorded manually or inferred from weight, reducing vehicle-capacity planning accuracy and causing inefficient loading. This paper presents a portable AI-assisted dimensioning prototype integrating a MaixSense A010 depth sensor, ESP32-S3 microcontroller, Bluetooth Low Energy communication, an Android driver application, and a YOLOv5n detection pipeline. The system captures RGB-depth data, detects parcel regions, and combines bounding-box geometry with depth information to estimate dimensions for driver decision support. Evaluation used a public box dataset with 2,426 images and 2,227 annotations. YOLOv5n achieved 99.5% mAP@0.5 and 67.7% mAP@0.5:0.95 after 10 epochs, indicating feasible lightweight parcel localization. System testing verified BLE discovery, connection, and data transfer. Practical constraints included a 46-minute battery runtime and reliance on cloud inference, highlighting improvements needed for robust field deployment. Future work should improve onboard processing, battery efficiency, and end-to-end measurement accuracy.

parcel dimensioning;RGB-D sensing;YOLOv5n;Bluetooth Low Energy (BLE);mobile logistics;computer vision;Android application;depth sensing;SDG 9 — Industry, Innovation and Infrastructure