Here’s a concrete marker strategy for AR navigation in a mall (optimized for real-world occlusion, long corridors, and re-acquisition).

1) Use 3 layers of references

Layer A — World anchors (primary, for accurate navigation)

Mount large AprilTag/ArUco tags on fixed, hard-to-move structures:

• Elevator cores (elevator bank centerline) — 1 set per core per floor
• Stairwell landing corners — 1 set per stair cluster per floor
• Main junction pillars/wayfinding pillars at corridor intersections
• Main entrances/lobby “origin” points (per floor, if applicable)

Goal: “world lock” so arrows/labels don’t drift onto the wrong storefront.

Layer B — Local anchors (secondary, for turn/door accuracy)

Mount medium tags near:

• Food court entrances
• Restroom zones
• Major store clusters where the user must make a turn
• Information desk/kiosk (optional but great for calibration)

Goal: keep turn guidance stable when global anchors are temporarily occluded.

Layer C — Marker-less fallback (must-have)

Enable SLAM/feature tracking so navigation continues when markers are hidden.

• Your UI should degrade gracefully (next section).

2) Tie markers to a state machine (so UX is safe)

Your app should behave differently based on tracking confidence:

• WORLD_LOCKED: marker detected with good quality → show precise step arrows, distance-to-turn.
• LOCAL_LOCKED: only local anchors visible or SLAM confidence moderate → show direction but reduce “snap/precision”.
• SLAM_ONLY: no markers visible → show high-level guidance (e.g., “approaching restroom area”) and avoid claiming exact alignment.
• RECOVERY: markers were lost → guide user to a nearby anchor zone (“look toward pillar signage”).

This prevents the classic mall problem: occluded tags, SLAM drift, and an arrow that “floats” incorrectly.

3) Placement rules that work in malls

A) Redundancy over density

• Put anchors in pairs/triads within a zone (e.g., two elevator-related anchors per elevator bank).
• Avoid “single points of failure” where the user must see exactly one tag.

B) Height & orientation

• Mount large anchors around 1.7–2.2 m height.
• Keep the tag plane facing the expected camera view direction (generally front-facing relative to where users walk).

C) Coverage logic (practical heuristic)

Per floor, ensure that along the main walking areas:

• Every major corridor section between junctions has at least one chance for a world anchor to be visible
• Every turn decision point has at least one local anchor opportunity

4) Concrete starting counts (per floor)

As a starting design before field testing:

• Large anchors: 6–12 per floor
• Medium anchors: 8–16 per floor
• Then add based on walk-test failures (don’t over-install everywhere up front).

5) Field test acceptance criteria (what to measure)

Perform a “walk-through" for each route and record:

• Re-lock time: how long it takes to return to WORLD_LOCKED after occlusion
• Arrow stability: jitter in position during 30–60 seconds of tracking
• Wrong precision rate: percentage of time you display “exact” overlays without being world-locked
• Detection reliability: success rate of marker detection by zone and distance

“Tracking type” in AR refers to how the glasses determine the user’s motion and the real-world coordinate system, so that virtual content stays locked to the correct positions.

1) SLAM tracking (Simultaneous Localization and Mapping)

Idea: The glasses create and update a map of the environment while estimating their own position within it.

How it works (high level)

• Use cameras (and often IMU) to detect visual features.
• Estimate the device pose (position + orientation) relative to previously seen features.
• Build/update a 3D map (point cloud, planes, sometimes meshes).
• Use the map to keep anchors stable.

Pros

• Works in many environments (especially those with texture/visual features).
• Doesn’t require pre-built markers or special setup.
• Can support larger spaces because it continuously builds the world model.

Cons/failure modes

• Can drift over time (though modern SLAM reduces this a lot).
• Performance drops in low light, motion blur, or low-texture scenes (blank walls).
• Changes in the environment can reduce stability.

2) Marker-based tracking

Idea: The system recognizes known “targets” (markers) in the scene and uses them to compute pose.

Common marker types:

• Visual fiducials: QR-like patterns, ArUco markers, AprilTags.
• Feature targets: printed images or special tracking cards.
• Visual anchor images: sometimes called image-based tracking (detecting reference pictures).
• UWB/Beacon-based (not visual, but “marker-like” anchors): device measures distance to known beacons.

How it works (high level)

• The camera detects the marker (or reference target).
• The glasses compute pose directly from the detected marker geometry.
• Virtual content is anchored to that marker’s known location.

Pros

• Often very stable and accurate as long as the marker is visible.
• Less computationally heavy than full SLAM in some setups.
• Great for repeatable tasks (training modules, industrial workstations).

Cons/failure modes

• Needs markers/targets present (or pre-installed environments).
• If the marker is occluded, moved, or out of view → tracking can fail or fall back.
• Less flexible for roaming/unknown environments unless you install lots of markers or use many reference targets.

3) When each is typically chosen

• SLAM: general-purpose AR, warehouses you move through, museums, outdoor/varied environments.
• Marker-based: factory assembly stations, labs, classrooms, “put this part here” workflows on fixed fixtures.

4) Hybrid approach (common in real products)

Many systems use both:

• Marker-based when targets are visible (stability/accuracy)
• SLAM when markers aren’t visible (coverage)

5) How to tell them apart in practice

• With SLAM, anchors generally “hold” as you move around, even without markers — until tracking starts to decline.
• With marker-based tracking, it works best when you can see the markers; occlusion or leaving the area often leads to noticeable tracking issues.