Measurement & Analytics
Dimensional Weighing & Logistics
Freight, parcel, and warehouse operations price and route shipments by dimensional weight, a number that’s only as good as the volumetric capture behind it. RealSense depth cameras generate point-cloud and aligned depth-to-color data fast enough to dimension irregular, non-cuboid packages moving on a conveyor or dock scale. Nothing has to stop, rotate, or get placed against a backstop.
Global-shutter variants capture every row of the frame at the same instant, so there’s no motion blur or skew from a rolling shutter on a moving belt. That’s what makes accuracy at line speed possible. And because the data feeds billing, slotting, and load planning through the same SDK used for static measurement, operators don’t maintain two stacks for stationary and in-motion capture.
Volume & Portion Analytics
Estimating volume from a 2D image is a losing proposition. Manual counts can’t quantify irregular, constantly changing shapes like scooped rice, ladled sauce, or shredded lettuce, so operators overprepare and waste a large percentage of what they buy.
RealSense depth sensors mounted above prep lines capture real depth data through glass hoods and display cases, locating target pixels and measuring distance to the food surface to calculate remaining volume in real time. Integrated RGB alignment lets vision AI classify the ingredient and its container at once, so one feed answers both “what is it” and “how much is left.” The result is inventory tracking accurate enough to trigger automated replenishment, cutting food cost and labor without adding headcount.
Anatomical & Body Measurement
Traditional fitting methods can’t keep pace with the variation in human anatomy or the sizing inconsistency across manufacturers. Brannock devices, tape measures, and sales-floor judgment all measure a foot one way and hope the size chart agrees.
RealSense stereo depth cameras capture high-resolution depth and color from multiple angles in a single scan under five seconds, building a millimeter-accurate 3D model of a foot, limb, or other body region without physical contact. Paired with an AI matching engine, that data recommends an exact size and shape match from a live product catalog. Retailers running this report double-digit conversion gains and meaningful reductions in size-driven returns
Capture in Motion
Most depth cameras are built for a subject that holds still. RealSense global-shutter models, including the D455 and the D550 series, are not. They expose every pixel at once instead of scanning row by row, so geometry stays accurate on a part moving down an inspection line, a package on a belt, or a limb that shifts mid-scan. A rolling shutter warps or blurs all three.
The same design keeps the depth and RGB streams in lockstep, which matters as much for classification as it does for motion tolerance. When depth and color are captured at the same instant, a model can trust that the volume it measured belongs to the object it identified.
Portability: Measurement That Goes to the Object
Not every measurement task can wait for the object to come to a fixed station. RealSense’s smallest cameras put dimensional accuracy in a handheld or field-mounted form factor. The D405 is 42 × 42 × 23 mm and about 60 g, smaller than a golf ball, and delivers sub-millimeter accuracy from 7 to 50 cm. That makes it viable for mobile inspection, wound measurement, and close-range quality checks a bench-mounted camera can’t reach.
The same open SDK and calibration model runs on a fixed overhead sensor and a handheld unit, so an application doesn’t get rebuilt from scratch when it moves from a stationary line to a technician’s hand.
Flexibility: From a Single Bench to a Networked Line
Measurement deployments rarely stay small or uniform, so RealSense doesn’t force every use case onto USB. GMSL/FAKRA interfaces on D400-series industrial variants give integrators automotive-grade, low-latency links up to 15 meters, with native sync across eight cameras, built for the noise and vibration of a factory floor.
Ethernet and PoE variants stream depth and color up to 100 meters, drawing power and data over one line, so a fleet of sensors goes onto existing network infrastructure without a parallel cabling project. Both run on the same SDK as USB cameras, so code written for one carries over.
Proven Accuracy, Built for the Field
A measurement system is only as good as its worst day on the floor. Spilled liquids, dust, vibration, glare, years of the kind of thing a lab test never accounts for. On-chip self-calibration holds depth accuracy stable across a fleet without a technician visiting each unit, and IP65-rated industrial housings resist dust ingress and projected water, so the same sensor runs on a warehouse floor, a production line, or an outdoor loading dock.
Volumental’s in-store scanners run non-stop through what a retail floor actually produces, spilled drinks included. PreciTaste mounts sensors above open food lines and behind glass hoods, reading depth through display cases and steam without dedicated shielding. Both operators cite reliability under messy conditions, not bench accuracy, as the reason the deployment scaled past a pilot.
Open-Source SDK for Custom Measurement Tools
RealSense measurement solutions are built on librealsense, the Apache 2.0–licensed, cross-platform SDK maintained on GitHub for the current D400 and D500 series depth cameras. Synthetic stream generation — point clouds, and depth aligned to color in either direction — gives developers the raw geometric data that volumetric and dimensional-weighing tools are built on top of, without writing a calibration or registration pipeline from scratch. The SDK repository ships working examples and wrapper integrations developers can build directly from:
- 3D reconstruction — wrappers/opencv/kinfu implements real-time volumetric reconstruction as a camera moves through a scene; examples/pointcloud and the wrappers/pcl integration cover point-cloud generation and further processing in the Point Cloud Library.
- Spatial analysis — examples/align and examples/align-advanced cover depth-to-color stream alignment and background segmentation built on it.
- Dimensional measurement — examples/measure shows the deprojection and calibration-intrinsics math (converting a pixel plus its depth value into a real-world 3D point) that dimensional-weighing and volumetric-measurement applications are built on. It’s published as a capability demo rather than a calibrated out-of-the-box measurement tool — production deployments layer additional filtering and calibration on top, the way Volumental and PreciTaste do in their own measurement pipelines.
Language and framework support includes C++, Python (pyrealsense2), C#/.NET, and Node.js, with integrations for ROS, ROS2, OpenCV, PCL, Unity, MATLAB, OpenNI, and Unreal Engine 4 — so measurement logic can be built directly into an existing robotics, inspection, or inventory stack rather than bolted on as a separate application.
