04 / 10 · Computer vision · PepsiCo · Stereo vision
Real-Time Volume Estimation
At PepsiCo, I derived a first-principles model for real-time volume calculation with no static reference markers. YOLO and OpenCV stereo vision estimate the volume of moving, partly hidden products, which enables dynamic pricing on mixed inventory.
- 0
- reference markers needed
- Real-time
- moving, occluded objects
Fig. 01
Try it
AI-generated voice and avatar
01
Problem & approach
The problem
A smart vending machine has to know which product, and how much of it, a customer took, without reference markers in view and while products move or hide behind each other.
The approach
A first-principles mathematical model for volumetric calculation, fed by YOLO detections and OpenCV stereo vision, so volume comes from depth instead of a reference object.
02
How it works
- Stereo pairTwo calibrated cameras watch the shelf from slightly different angles.
- DetectA fine-tuned YOLO model detects and classifies each product in the frame.
- DepthOpenCV stereo vision turns the pair of images into depth, so every detection gets a distance.
- VolumeA first-principles model computes volume in real time, even for moving and partly occluded products, with no static reference markers. That makes dynamic pricing on mixed inventory possible.
03
Results & stack
- 0
- reference markers needed
- Real-time
- moving, occluded objects
- Python
- C++
- YOLO
- OpenCV
- Stereo vision