Quick answer: Occlusion in augmented reality is what happens when a real object correctly blocks part or all of a virtual one from view, the way a real coffee cup in front of a virtual lamp should hide the part of the lamp sitting behind it. Get this wrong and virtual content floats on top of everything like a sticker, even when it is supposed to sit behind your hand or your sofa. Getting it right is the clearest sign that an AR system understands the room rather than just painting over it.
This entry covers how AR systems figure out what is in front of what, where occlusion shows up in apps you already use, how it differs from the tracking that keeps virtual objects anchored in place, and what an AR scene looks like when occlusion fails.
What Is Occlusion in AR?
In ordinary computer graphics, occlusion, also called hidden-surface determination, just means figuring out which surfaces of a 3D scene are visible from the camera’s position and which are blocked by something closer. AR adds a harder version of that problem: virtual objects also have to be correctly hidden behind real ones, and the real world is not geometry the app already has stored.
A phone placing a virtual chair in your living room does not have a 3D model of your coffee table sitting in front of that chair. It has a camera feed and whatever the AR framework can infer about the room’s depth and shape in real time. Occlusion is the piece of the system responsible for using that inference to draw the chair correctly, cut off wherever your table should hide it, rather than painted whole on top of the video.
How Does Occlusion Work?
Most AR occlusion starts with a depth map, an estimate of how far away every pixel in the camera view is. Once the system knows the distance to the real world and the distance to the virtual object it is placing, it compares the two per pixel and only draws virtual pixels that are closer than the real surface in front of them, the same depth-testing logic used in ordinary 3D rendering for decades.
Getting that depth map is the hard part. Frameworks like Google’s ARCore Depth API estimate it from a single moving camera using structure from motion, tracking how points shift between frames to infer distance. Those estimates hold up best when the device is roughly half a meter to five meters from what it is looking at, which covers most of what a phone app is asked to do and little else.
Devices with a dedicated depth sensor skip most of that inference. Apple added a LiDAR Scanner, which measures distance by timing reflected light, to the fourth-generation iPad Pro in March 2020, and has since built it into several iPhone Pro models.
A LiDAR sensor gives AR software a direct distance reading instead of one estimated from motion, which makes occlusion faster to compute and far more reliable on plain, textureless surfaces where motion-based depth tends to struggle. That same depth data is often what a framework uses to build a 3D mesh of the room, which occlusion then checks virtual content against.
Occluding real people is its own problem, since a person moves independently of the room around them. Apple’s ARKit handles this with person segmentation, a machine learning model that identifies which pixels in the camera feed belong to a human, combined with depth so virtual content is hidden only where a person is actually closer to the camera than the virtual object. That feature requires an A12 chip or later and has been available since iOS 13.
Where Is Occlusion Used?
Furniture apps depend on it constantly. A virtual sofa that a shopping app places in a room needs to disappear behind a real doorway or a person walking past it, or the preview stops looking believable the moment anything moves between the camera and the object.
Mixed reality headsets lean on occlusion even more heavily than phone apps, since a virtual object sitting close to the viewer’s hands or a real desk needs to respond to that geometry constantly rather than once at placement. It is one of the features that separates devices like Microsoft’s HoloLens and Apple’s Vision Pro, both usually described as mixed reality hardware, from AR that simply overlays graphics on a flat video feed.
Occlusion also shows up in AR navigation and industrial tools, where a direction arrow or a repair instruction needs to appear to sit on the actual street or the actual piece of equipment rather than hover in front of it regardless of what is nearby.
Occlusion vs World Tracking: What’s the Difference?
It is easy to conflate occlusion with world tracking, the SLAM-based process that figures out where the device is and keeps virtual content anchored to a fixed point in space, but they solve different problems. World tracking answers where a virtual object should sit and keeps it from drifting as you move. Occlusion answers what should be visible from there, given everything else in the scene.
A system can have excellent world tracking and no occlusion at all, which is exactly what early phone-based AR looked like: objects stayed correctly anchored in place but were always drawn on top of everything, real hands included. The two capabilities usually ship together in modern frameworks, but they rely on different inputs, position and motion data for tracking, depth and segmentation for occlusion, and each can fail independently of the other.
What Breaks Without Occlusion?
Without it, an AR scene reads as a flat overlay rather than a real layer of the room. A virtual object that should be partly hidden behind a table or a hand instead renders as a solid shape stamped on top of the video, which is one of the fastest ways to break the sense that the content belongs in the space at all.
The effect compounds with distance and clutter. A single virtual object in an empty room can get away with weak or missing occlusion; a busy scene with several real objects at different depths exposes it immediately, since every one of them needs to interact correctly with the virtual content for the illusion to hold.
That is also why occlusion quality tends to track hardware quality closely. Phones without a dedicated depth sensor can still produce reasonable occlusion using motion-based estimation and person segmentation, but a LiDAR-equipped device produces cleaner edges and holds up better in low light or against plain surfaces, which is a large part of why LiDAR became a selling point for AR-heavy apps rather than a niche spec.