Portable vision concept

Point. Walk. Count.

RecycleScan Handheld · one self-contained scanning unit

01 / Walk around the batchSIMULATION · READY
Camera + optional depth
Jetson + cooling
Protected battery
Touchscreen
01 / PREPARESpread the materials

Separate touching items. Keep the batch still during scanning.

02 / WALK & SCANLook from several sides

One spatial identity per item keeps repeat views from adding counts.

03 / CHECKReview each material

Check missed objects and uncertain matches before accepting a batch.

A handheld recycling counter you can demonstrate

The operator holds a single enclosure containing a camera, Jetson computer, screen and battery. As they move around a stationary batch, the camera sees new sides. The screen groups detected items into plastic bottles, cans, glass bottles, foam pieces and cartons.

1 · SeeCapture a sharp frame and segment individual visible objects.
2 · LocateEstimate camera pose and object locations using calibrated depth / visual-inertial tracking.
3 · MatchAssociate each observation with an existing item, or create a new identity.
4 · CountCount unique accepted identities. Review ambiguities; do not add frame totals.

Why spreading the pile matters

The green viewing rays are a schematic aiming aid, not a calibrated camera frustum. The demo uses actual 3D lines of sight to test synthetic sample points on each object. Walking reveals some obscured items. Spreading the batch reduces occlusion further. When the arrangement changes, the count resets so the next scan represents the new layout.

This is an explanatory simulation, not working ML or SLAM. Perfect item identities come from the simulated scene. Real duplicate prevention is a difficult engineering task, especially with identical bottles, reflective cans, clear glass or objects moved during scanning. A depth sensor alone does not solve identity matching.

What “Styrofoam” means here

The foam class represents EPS foam pieces such as packaging and cups. Count one separated piece as one item. A camera cannot certify polymer chemistry or local recycling eligibility; foam may be accepted by weight, volume or at specific facilities.

Build the real model

Label your own recycling images with individual masks in Roboflow or another annotation tool. Keep training and test batches separate. Validate per-class count error and identity switches across handheld videos. Add a “needs review” state when a track is uncertain. Start with one spread-out tray and one approved overhead photo before attempting free-moving spatial counting.

Roboflow tracking guide · Jetson inference deployment

All inside one housing

Concept enclosure: approximately 220 × 140 × 85 mm, plus a grip. This is a packaging target, not a verified mechanical fit. A compact carrier is needed for a refined handheld product; check the complete board, connectors, heatsink, battery and cable bend space in CAD.

PartConnection / purpose
RGB cameraSupported CSI module and matching cable, or USB camera → Jetson. Forward/downward view.
Depth + IMU optionSupported USB sensor → Jetson. Calibrated pose/depth aid spatial matching. Verify runtime and drivers.
TouchscreenCompatible display interface plus USB touch → carrier. Adapter may be required.
PowerProtected pack → fuse near pack → switch → regulated supply → exact carrier DC input.
CoolingHeatsink + fan with unobstructed inlet and outlet; isolate warm hardware from the hand.
StorageSupported SSD / microSD stores model, batch records and optional review images.

Handheld design details

Put the battery low and near the grip for balance. Add a wrist strap, recessed camera window, removable service cover, clear fan vents and a thumb trigger. Keep screen glare low. Measure mass, grip comfort, surface temperature and runtime on the assembled prototype.

Use a protected pack and its matched charger. Confirm voltage, polarity, peak current, fuse and connector ratings from the exact carrier documentation. Do not use GPIO as a power input or assume Xavier NX and Orin Nano have interchangeable carriers. Shut down before disconnecting power.

NVIDIA Orin Nano hardware layout · Supported hardware · Find Xavier NX carrier documentation

Planning bill of materials · CAD

Editable allowances, not supplier quotes. One of each. Excludes labour, tax, shipping and model/software fees. Mass values are placeholders for packaging estimates.

ComponentCADGrams

RecycleScan Handheld Recycling Counter

Explore a handheld recycling counter concept with synthetic spatial scans, tracked item identities, runtime estimates, and BOM exports.

What this tool does

Explore a handheld recycling counter concept with synthetic spatial scans, tracked item identities, runtime estimates, and BOM exports.

Who this is for

How to use this tool

  1. Review the controls and instructions on the page
  2. Enter your information or choose the options you need
  3. Review the result in the app workspace
  4. Export, download, or save the result when the tool provides that option

Inputs

Inputs vary by tool and are provided through the visible fields, controls, files, or interactive workspace on this page.

Outputs

Outputs depend on the selected options and may include an analysis, generated content, visualization, calculation, simulation, or downloadable file.

Example use cases

Explore more SolvProb tools

Frequently asked questions

Is this tool free to use?

Yes. This SolvProb browser tool is available to use without installing a desktop application.

Do I need to download software?

No. Use a modern browser with JavaScript enabled. Tool-specific browser requirements are shown in the app when applicable.

Can SolvProb build a custom version?

Yes. SolvProb can adapt a tool or build a custom website, app, automation, or workflow for a specific business need.

Need a custom tool built for your business?

SolvProb builds custom web apps, automation tools, and interactive experiences tailored to your workflow.

Talk to SolvProb →