Counting parts or boxes on a moving conveyor belt with computer vision

Problem: Building a reliable computer vision application to count fast-moving objects

In manufacturing and logistics settings, accurately counting items on a moving conveyor belt is essential for throughput tracking, quality assurance, and process automation. But building a custom computer vision model from scratch is often time-consuming and inflexible — models can break when product sizes vary, lighting shifts, or environmental factors like glare and dust interfere. Scaling such systems across multiple lines or facilities only adds complexity.

Developers tasked with improving these systems usually face two options: implement brittle sensor logic or go down the ML rabbit hole — collecting image data, annotating datasets, and training and maintaining computer vision models. That’s a nonstarter for most teams without dedicated ML expertise.

‍

Solution: Groundlight's computer vision tools to build a custom counting model for moving parts

Groundlight AI offers computer vision tools for developers to build robust, vision-based counting systems in minutes using simple natural language prompts and a Python SDK — no model training and maintenance needed.

Here's how you use Groundlight to deploy a computer vision model to count moving parts:

• Connect any IP camera or image stream to the Groundlight API
• Point the camera to the objects that are moving on the conveyor belt
• Install Groundlight Edge to ensure rapid object counting
  
• Use a natural language query such as "Count the number of packages on the conveyor"
  
• Handle results programmatically in your Python code with one API call
  
• Add optional logic for thresholds, alerts, or integrations with existing systems

The Groundlight platform handles vision model orchestration, error correction, and continuous improvement behind the scenes — so developers can focus on building reliable automation logic, not ML infrastructure.

‍

Results: a reliable computer vision model for fast-moving objects

Groundlight takes care of the labeling for you with its "Cloud Labelers" or humans-in-the-loop - however, if you want to review or modify results, it's easy to do so.

Teams using Groundlight to build conveyor-based counting applications report:

• Quick setup — functional vision counting logic in under an hour using just Python
  
• High accuracy — robust to motion blur, changes in lighting, and other changes in the environment
  
• Less maintenance — no model tuning, retraining, or deployment overhead
  
• Easy integration — simple to plug into existing control systems, dashboards, or data pipelines
  

Whether you're building a factory dashboard, a quality control alerting system, or just need a drop-in replacement for fragile sensor logic, Groundlight gives you the power of computer vision with the simplicity of a few lines of code.