AI-Powered Load Counting App

I already have IP cameras installed at the loading bay and now need a software solution that watches those live feeds and tells me—accurately and in real time—how many items go on and off each lorry. The system must recognise three categories: Bags, Boxes and Tins. People sometimes carry two bags at once, so the model has to distinguish single-bag from double-bag events and still keep a reliable tally. Boxes and tins also come through the same point, each in more than one size, so the detection logic has to cope with differing dimensions rather than relying on a fixed template.

Here’s the workflow I have in mind:
• Your application pulls the RTSP/HTTP stream from my existing IP cameras.
• A computer-vision model (OpenCV, TensorFlow, YOLO or similar) detects the item type, counts it, and determines the direction of movement—onto or off the truck.
• Counts are logged with time-stamps and can be viewed on a simple web dashboard and exported as CSV.
• If the camera loses connection or the count confidence falls below a threshold, the system raises an on-screen alert so we can check manually.

I need everything packaged so I can run it on a Windows PC beside the loading point—installers, model weights, source code and brief setup documentation included. Delivery deadline is within a month from project start, with at least one interim demo so we can confirm detection accuracy against live footage.