Vision-Based Sorting Conveyor

Context

A benchtop demonstration line: parts ride a conveyor past a camera, get classified by color and outline, and a pneumatic gate diverts each part into the right bin.

My role

Built the vision pipeline and the handshake between the vision computer and the PLC controlling the conveyor and gate.

Constraints & tradeoffs

Classification runs classical OpenCV (color thresholds plus contour features) rather than a neural network — the part set is small and controlled, and deterministic thresholds are easier to debug on the line than a model.

Approach

A Raspberry Pi runs the OpenCV pipeline and signals the PLC over discrete I/O; the PLC owns all motion and timing so a vision hiccup can never leave the gate in an unsafe state.

Result

Sorts the demo part set reliably at conveyor speed. Placeholder metrics — to be replaced with real ones.