This project focused on developing a computer vision–based system to automatically count boxes unloaded from a delivery truck during warehouse operations. The objective was to replace manual counting with a reliable visual workflow that operates during real unloading conditions, including variable lighting, partial occlusions, and changing backgrounds.

A configurable region of interest (ROI) was defined across the unloading conveyor to isolate the counting zone and prevent false detections from surrounding activity. A detection-based vision model was used to identify boxes as they passed through the ROI. The system tracked object crossings through a virtual counting line and incremented counts only when a box fully crossed the defined boundary, preventing duplicate or missed counts.

The solution generated a real-time visual overlay showing the ROI, counting line, detected objects, and current count value, enabling operators to easily verify system behavior. The counting results can be logged for operational reporting and inventory reconciliation.

This automated workflow reduces human error, improves unloading throughput, and provides consistent, auditable box counts without requiring changes to existing material handling equipment.