Home / Applications

Applications

Three industries, one approach: get the light and the capture right, then put the model on the hardware in the field. Every number comes with its test conditions.

Research

Coffee Bean Sorting

Small roasters still pick out defective beans by hand. Our goal is a sorter small roasters can afford: first count and track every bean accurately on the conveyor, then judge defects with multispectral imaging.

Approach

  • A lightweight detector running in real time on a Raspberry Pi 5
  • Global motion-compensated tracking so no bean is counted twice
  • Multispectral capture to reveal defects RGB cannot separate
93%Conveyor counting accuracy370 of 399 beans counted correctly in the test video
Raspberry Pi 5Real-time at the edge, no PC or cloud
Co-developed with an equipment maker

Medication Recognition

An automatic dispenser needs to know two things: is there a pill in the tray cell, and where should the nozzle pick it up? We develop both vision steps with the equipment maker, running on a small computer inside the machine.

Approach

  • Presence check: compare brightness under two lighting stages and subtract an empty-tray reference
  • Pick point: instance segmentation outlines each pill, then computes the best suction point
  • All models and libraries under commercially usable licenses
36 / 36Pick points correct in the operating-scenario testInternal test, confidence threshold 0.7
Raspberry Pi CM4Inside the machine, fully offline
In store trials

Skin Analysis

An ordinary camera records only red, green and blue. The DotData Skin Mirror splits 400–950 nm into multiple bands and, using how melanin and hemoglobin absorb each wavelength, turns spots, brown areas and red areas into scores and overlay photos an advisor can explain.

Privacy by design

  • Face photos stay on the in-store host; nothing is kept on the phone
  • Photos and client nicknames are deleted when the retention period ends
  • No ads or tracking in the app

Results describe skin appearance and are not a medical diagnosis.

Violet-blue

400–480 nm
400500600700800950 nm
Past the dashed line at about 700 nm, the human eye sees nothing.

What it shows
In the report

Facing a similar problem?

Send sample photos or a description of your site. We will reply with an initial assessment: which light, which camera, and roughly what it will cost.

Email
Vision assessmentCo-developmentStore trial