Depth data only proves its value once it reaches the board running inference, and that handoff is where a large share of integration time gets lost. The Orbbec Femto Mega, an edge AI depth camera built around precisely this handoff, pairs an indirect Time-of-Flight (iToF) sensor with onboard NVIDIA Jetson processing, then streams depth, RGB, and point cloud data over a single USB-C or Ethernet connection to whichever host runs the inference stack — including a Jetson Orin developer kit, NVIDIA's edge AI computing platform for embedded and robotics workloads. This guide walks through what that connection looks like in practice, from cabling to SDK output, for developers wiring the Femto Mega into a Jetson Orin–based workflow.

Why iToF fits indoor edge AI workflows

Indirect Time-of-Flight measures the phase shift of modulated infrared light reflected off a scene, while stereo vision triangulates depth between two image sensors. The distinction matters indoors: iToF returns a full depth frame from a single sensor head without a wide stereo baseline, which keeps the camera compact and avoids the correlation errors stereo systems can produce on low-texture or repetitive surfaces.

The Femto Mega's depth sensor uses an 850nm infrared wavelength and outputs resolution up to 1024×1024 at 15 fps in wide field-of-view mode — a 120-by-120-degree field — or 640×576 at 30 fps in narrow field-of-view mode. Working depth range runs roughly 0.25 to 5.46 meters depending on the selected mode.

That range-and-resolution profile suits the indoor perception tasks edge AI pipelines are usually built for. Quality inspection at close range, dimensioning and volumetric capture in logistics, patient positioning, and guided workout or motion capture applications all sit within a few meters of the sensor. Orbbec's own product documentation lists the operating environment as indoor use at 10°C to 25°C and 8 to 90 percent relative humidity, non-condensing. Wide-temperature-range or outdoor operation calls for a different camera in Orbbec's lineup, so this integration path assumes a controlled indoor environment — the setting where iToF depth data holds up best.

The Femto Mega also carries an integrated inertial measurement unit (IMU), so motion data can be correlated with each depth frame. For autonomous mobile robot (AMR)-adjacent or handheld capture use cases, the IMU stream lets downstream inference code compensate for sensor movement between frames instead of assuming a static camera.

What the physical connection actually looks like

The Femto Mega ships with two data connection options: USB 3.0 Type-C or Ethernet with Power over Ethernet Plus (PoE+). Both paths carry the same processed depth, RGB, and IMU output the difference is cabling distance and power delivery, not data format.

For a bench-level or robot-mounted setup where the Jetson Orin host sits within a meter or two of the camera, USB 3.0 Type-C is the simpler path. Power draw is low — the camera is specified at an average of under 13W — so a single Type-C cable to a powered USB 3.0 port on the Jetson Orin carrier board handles both data and power. Orbbec also lists a DC 12V/2A power adapter and PoE+ as alternate power inputs, which work better when the camera sits farther from the host: an overhead-mounted unit in a quality inspection cell, for example, or a fixed camera watching a logistics dimensioning station. In a PoE+ deployment, a single Ethernet run to a PoE+ injector or switch near the Jetson Orin host carries power and data together, simplifying cable routing.

Confirm the physical mount matches the fixture before wiring anything up. The Femto Mega uses a ¼-20 UNC mount on the bottom and four M2.5 mounting points on the side; a standard camera mount or a custom bracket with the correct hole pattern will both work, depending on how the rest of the rig is built.

One architectural point is worth stating plainly. The Femto Mega's onboard NVIDIA Jetson Nano handles the camera's own depth engine processing, converting raw ToF returns into the depth map, point cloud, and IR/RGB output the SDK exposes. The Jetson Orin in this setup is a separate, external host device: it receives the already-processed data over USB or Ethernet and runs the downstream inference workload, whether that is object detection, pose estimation, or a custom model. The camera does not run the inference graph — it hands the Orin-based host clean, calibrated depth and RGB streams to work from, freeing the host's compute budget for the actual AI workload instead of raw sensor processing.

Bringing point cloud data into the pipeline with the Orbbec SDK

Once the camera is connected and powered, the Orbbec SDK turns the USB or Ethernet stream into usable frames on the host side, exposing point cloud, depth map, IR, and RGB as distinct outputs. A typical Jetson Orin integration pulls these through the same general sequence regardless of connection type.

First, the SDK enumerates connected devices and opens a pipeline to the Femto Mega, selecting the depth and color stream profiles matched to the chosen field-of-view mode.

Second, depth-to-color alignment is configured so the depth and RGB frames share a common coordinate space. That matters for any downstream step that needs to associate a depth value with a specific pixel in the color image, such as segmenting an object and then measuring it.

Third, the SDK's point cloud generation step converts the aligned depth frame into a 3D point cloud — the format most inference and perception pipelines expect for object localization, volume estimation, or obstacle mapping.

From there, the point cloud or aligned depth and RGB frames are handed off to whatever inference framework the Jetson Orin host runs. The SDK is cross-platform and actively maintained as open-source SDK v2, with ROS and ROS 2 wrappers — bindings for the Robot Operating System common in robotics stacks — and Python support, so the handoff typically fits into an existing robotics or computer vision codebase rather than requiring a custom bridge layer.

Teams migrating from an Azure Kinect DK setup can use Orbbec's K4A Wrapper, which lets Femto Mega code paths mirror the Azure Kinect Sensor SDK — easing the migration instead of requiring a rewrite against a new API from scratch.

The camera also supports hardware trigger for multi-sensor synchronization. Build this into pipeline design early if the deployment involves more than one Femto Mega, or if it needs to stay frame-aligned with another sensor on the same platform. Handling sync at the trigger level, rather than reconciling timestamps after the fact in software, keeps the point cloud data cleaner going into inference.

Practical checklist before you start

Before writing integration code, nail down four decisions:

1. Which connection type fits the cable run and power budget?
2. Which depth mode — wide or narrow field of view — matches the working distance and required frame rate?
3. Does the deployment need hardware trigger sync across multiple sensors?
4. Which SDK language binding — C++, Python, or a ROS wrapper — matches how the rest of the Jetson Orin pipeline is built?

Getting these four decisions right before the first line of integration code avoids the most common source of rework: discovering partway through development that the selected depth mode does not cover the working distance the application actually needs.

Frequently asked questions

Does the Femto Mega run its own AI inference, or does that happen on the Jetson Orin host?

The camera's onboard NVIDIA Jetson Nano handles depth engine processing only. It converts raw sensor data into depth maps, point clouds, and aligned RGB output. The inference model runs on the separate host device — in this guide, a Jetson Orin platform receiving that processed data over USB or Ethernet.

Can Ethernet and USB 3.0 be used at the same time for redundancy?

The product specification lists USB 3.0 Type-C and Ethernet as alternate data connection options. The Femto Mega is designed around a single active data connection at a time rather than simultaneous dual-stream operation, so the architecture should be planned around one connection type per camera.

What depth mode should be chosen for a close-range inspection task versus a wider-area logistics task?

Narrow field-of-view mode delivers 640×576 depth resolution at 30 fps, which suits close, detail-oriented tasks like quality inspection, where frame rate and update speed matter more than coverage area. Wide field-of-view mode delivers the full 1024×1024 resolution at 15 fps across a much larger 120-by-120-degree field, which fits dimensioning or scene-level tasks where covering more of the space matters more than frame rate.

Is the Femto Mega rated for outdoor or variable-temperature deployment?

No. Orbbec's specification lists the operating environment as indoor use within a 10°C to 25°C range, so this integration path is built around controlled indoor conditions rather than outdoor or wide-temperature-swing environments.

Where to go next

The steps above cover the core integration path, but depth mode selection, exact SDK API calls, and current firmware behavior are worth confirming directly against the source before a is finalized. The Femto Mega product page carries the full specification table and datasheet, while the Orbbec SDK documentation covers the current API reference for point cloud generation, stream alignment, and the ROS/ROS 2 wrappers referenced here.

This guide originally appeared on FinTechZoom.