If you’re looking at a structured light 3D camera for computer vision and wondering how it’s different from a Time-of-Flight camera, you’re asking the right question before you buy hardware, not after. Both types of camera give you a depth map, a picture where each pixel carries a distance value instead of just color, but they get there through completely different physical processes. That difference shows up directly in how accurate the camera is, how far it can see, and how well it holds up once you move it out of a controlled lab environment. This guide walks through both technologies in plain terms, where each one wins, and where the Astra series specifically fits for developers building gesture recognition, 3D scanning, or pose estimation projects.
How structured light actually works
A structured light camera projects a known pattern of infrared dots or lines onto whatever it’s looking at, using an infrared projector built into the device. A separate infrared camera, placed a fixed distance away from the projector, then looks at how that pattern bends and shifts as it lands on surfaces at different depths. That fixed distance between the projector and the camera is called the baseline, and it’s the same basic geometric idea stereo cameras use with two lenses, except here one side of the pair is a projector instead of a second camera. The system already knows what the pattern should look like on a flat surface, so it calculates depth by measuring how far each part of the pattern has shifted from that expected position, a process called triangulation.
Orbbec’s Astra 2, the flagship of the Astra series, works exactly this way. It uses a 75mm baseline and Orbbec’s own custom processing chip, the MX6600, to handle that pattern-matching calculation inside the camera itself rather than pushing the work to your computer’s CPU.
How ToF works, briefly
Time-of-Flight cameras solve the same problem differently. Instead of projecting a pattern and measuring how it deforms, a ToF camera sends out a pulse of infrared light and measures how long it takes to bounce back from the scene, pixel by pixel, across the whole sensor at once. There’s no baseline involved and no pattern to interpret. It’s a more direct measurement: how long did the light take, and how far did it therefore travel. Orbbec’s Femto Bolt is a ToF camera built this way, and it’s worth knowing about mainly as the contrast case for everything below.
Where structured light has the edge
The first thing structured light does well is short-to-mid range accuracy. Because the depth calculation comes from resolving fine detail in a projected pattern, the camera can pick up small variations in surface shape with real precision, as long as the object is within its working range. Astra 2’s typical depth accuracy is 0.16% at 1 meter and 0.3% at 2 meters, tightening noticeably the closer you get to the camera. Its overall depth range runs from 0.6 meters to 8 meters, with an ideal working range of 0.6 to 5 meters where that accuracy holds up best.
The second advantage is indoor, controlled-lighting environments. Structured light depends on the camera being able to see its own projected pattern clearly against the background. Indoors, under normal room lighting, that’s a solved problem. Outdoors in bright sunlight, the sun’s own infrared output can wash out the projected pattern, which is why structured light cameras like the Astra series are built and positioned for indoor and semi-outdoor use rather than direct outdoor deployment. If your project lives indoors, in a lab, a clinic, a retail space, or a studio, this isn’t a limitation you’ll ever run into.
The third thing worth knowing is that this precision comes with genuinely dense output. Astra 2 delivers depth resolution up to 1600×1200 pixels at 30 frames per second, which means a lot of fine spatial detail per frame, useful for anything where the shape of an object or a person’s body actually matters, not just their rough position in space.
Where ToF has the edge
ToF cameras generally win on compactness and field of view. Because there’s no baseline geometry to fit into the housing, a ToF sensor can be built into a smaller, more self-contained unit while still covering a wide angle of the scene. They also tend to hold up more evenly across a longer working distance, since every pixel measures its own independent light travel time rather than depending on how well a projected pattern happens to land at a given distance. If your application needs a small camera covering a wide scene from a single vantage point, or needs to work reliably from very close range out to several meters without the accuracy dropping off sharply, ToF is usually the better starting point.
Common use cases for structured light
Gesture recognition is a natural fit because the interaction typically happens within a meter or two of the camera, indoors, exactly where structured light’s accuracy is strongest. The dense depth output also makes it easier to track the fine hand and finger movement gesture recognition depends on.
3D scanning is where structured light’s precision really pays off. Whether you’re scanning a small object for reverse engineering, capturing a product for e-commerce, or building a full-body scan for fitness or healthcare applications, the sub-1% depth accuracy at close range translates directly into a scan that actually matches the real object’s dimensions, not just a rough approximation of its shape.
Human pose estimation benefits from the same density and accuracy. Estimating joint positions and body posture requires the camera to resolve real detail across a person’s body at typical indoor distances, and Astra 2’s combination of depth resolution and near-range accuracy gives a pose estimation pipeline more usable signal to work with than a lower-resolution or longer-range-optimized sensor would.
Getting started without over-engineering it
If you’re new to depth cameras, the fastest way to get moving is the Orbbec SDK, which is open source, cross-platform across Windows, Ubuntu, macOS, and Android, and includes ROS and ROS2 wrappers if your project is robotics-adjacent, along with Python support if it isn’t. You don’t need to hand-roll your own depth processing pipeline. The camera and SDK handle the pattern projection, the triangulation math, and the depth-to-color alignment before the data ever reaches your application code, so your actual development work starts at the point cloud or depth map, not at the sensor physics.
A few questions people new to this ask
Is one of these technologies just better than the other? No, and that’s the whole point of understanding the difference before you buy. Structured light wins on close-range precision and dense detail indoors. ToF wins on compactness and consistent performance across a wider range. Neither is a universal upgrade over the other, they’re suited to different jobs.
Can I use a structured light camera outdoors? You can, in indirect or shaded conditions, but direct sunlight will degrade its accuracy because the sun’s infrared output competes with the camera’s own projected pattern. If your project needs to work reliably in bright outdoor light, structured light isn’t the right technology regardless of which specific camera you pick.
Do I need to understand the math behind triangulation to build with these cameras? No. The camera and its SDK handle the physics. Understanding the basic mechanism helps you predict how the camera will behave in your specific environment, close range versus far, indoors versus outdoors, single object versus full scene, but you don’t need to implement any of the underlying calculation yourself.
Which one should I start with if I’m just experimenting? If your project is indoors and involves people, hands, or objects at typical desk or room distances, gesture recognition, pose estimation, or small-object scanning, start with a structured light camera like the Astra series. If you need a compact sensor covering a wider scene or a longer variable range, look at a ToF option instead.
The short version: structured light gets you precision and detail at the distances most indoor computer vision projects actually operate at, and it does that by solving a geometry problem rather than a timing problem. Once you know which problem your application actually has, range versus precision, indoor versus outdoor, compact versus dense, the right camera technology tends to pick itself.
Lynn Martelli is an editor at Readability. She received her MFA in Creative Writing from Antioch University and has worked as an editor for over 10 years. Lynn has edited a wide variety of books, including fiction, non-fiction, memoirs, and more. In her free time, Lynn enjoys reading, writing, and spending time with her family and friends.
