# Radar, Camera or Depth Sensor? Choosing the Sensing Layer for an Interactive Installation

> Every interactive installation needs a way to know someone is there. Radar detects presence and approach in any lighting and even through non-metallic surfaces. Depth sensors give reliable skeleton tracking at close range. Cameras with vision models read gesture, object and appearance but need light and raise privacy questions. Most production installations combine two.

**Published:** 2026-08-20 - **By:** MLABART

## The question that decides the sensor

Do not start from the sensor. Start from one question: what is the minimum the installation actually needs to know?

Each level up costs more, fails in more ways and takes longer to calibrate. Choosing the level above what the interaction needs is the most common and most expensive sensing mistake.

- Is anyone there? -- presence only. Radar or a simple beam is enough.
- Where are they? -- position in a zone. Radar, LiDAR or depth.
- What is their body doing? -- skeleton and gesture. Depth sensor, or a vision model on a camera.
- Who or what is it? -- face, age bracket, clothing, object, product held. Camera plus a vision model, and a privacy decision.
- Did they touch it? -- contact. Touch foil, capacitive surface or a physical control.

## The four sensing families, compared

This is the comparison most briefs need and rarely contain:

| Sensing | Works in darkness | Range | Reads skeleton / identity | Privacy profile | Typical failure on site |
| --- | --- | --- | --- | --- | --- |
| mmWave radar | Yes | Metres, adjustable zones | Presence and position; no identity | Strong -- no image captured | Metal structure and moving crowds create ghost targets |
| Depth sensor (ToF / structured light) | Yes | Short, roughly room-scale | Skeleton tracking, yes | Moderate -- depth map, not photo | Direct sunlight and infrared stage lighting wash out the signal |
| RGB camera + vision model | No -- needs light | Wide, lens-dependent | Gesture, object, appearance | Weakest -- must be designed for | Backlight, crowd occlusion, high-contrast venue lighting |
| Touch / capacitive / physical control | Yes | Contact only | No | Strong | Wear, moisture, and glove or long-nail contact failures |

## Why most installations use two sensors, not one

A single-sensor design is fragile. The pattern that survives real venues is layered: a robust, low-resolution sensor decides that something is happening, and a high-resolution sensor handles the detail once the guest is in the right place.

A typical layering: radar wakes the piece and detects approach from several metres away, then a depth sensor takes over for gesture tracking once the guest is inside the interaction zone. The radar never fails in bright or dark conditions; the depth sensor only has to work in a small, controlled volume where you can shade it.

This layering also solves attract mode. The piece knows the difference between an empty room, someone walking past, and someone who has stopped -- and can behave differently in each state instead of looping the same invitation forever.

## Field conditions that break sensing

These are the conditions that cause commissioning-week emergencies. Check them during the site survey, not on install day:

- Direct or reflected sunlight -- atriums, glass roofs and shopfronts wash out infrared depth sensors completely at certain hours of the day. The failure is time-of-day dependent, so a morning site visit can miss it.
- Stage and event lighting -- moving heads and infrared-heavy fixtures interfere with depth sensing; the piece works in rehearsal and fails when the lighting cue fires.
- Reflective and glossy floors -- polished mall floors create mirror targets that both radar and vision models will happily track as a second person.
- Glass and mesh -- visually transparent, but they change how radar and depth behave. Radar can be an advantage here: it can see through a non-metallic decorative skin, letting you hide the sensor entirely.
- Crowd occlusion -- a sensor calibrated with one person in an empty space behaves differently when twenty people are queueing. Test with bodies, not with an empty room.
- Dark clothing and unusual silhouettes -- infrared-absorbing fabrics and large coats degrade depth tracking. Winter events are measurably harder than summer ones.

## Privacy: decide before you build, not after

Any camera-based installation in a public or retail space needs a stated position on privacy, and it is far cheaper to design in than to retrofit after a client legal review.

The defensible default: run inference on the local machine, keep nothing after the interaction ends, never write frames to disk, and post clear signage. If a guest takeaway requires storing an image, make it opt-in, tie deletion to a stated window, and put that window in the contract.

This is also a selection argument. If the interaction only needs presence and position, radar delivers it without any image ever existing -- which removes the entire question rather than answering it.

## Matching sensing to the installation type

How the choice usually lands in practice:

| Installation | Primary sensing | Why |
| --- | --- | --- |
| Large interactive wall or floor, high footfall | Radar, or radar plus depth | Robust in any light, multi-person, no privacy exposure |
| Gesture-driven game or sport interaction | Depth sensor | Skeleton tracking is the requirement; the zone can be shaded and controlled |
| AI photo or generative portrait | RGB camera plus vision model | The image is the product; privacy handling must be explicit |
| Product-recognition or pick-up interaction | Camera, or sensor in the fixture | Needs to know which object, not just that someone is there |
| Mechanical piece triggered by approach | Radar behind the decorative skin | Sensor stays invisible; motion is safety-critical so false triggers must be low |


## FAQ

### What is the difference between radar interaction and camera interaction?

Radar detects presence, distance and movement without producing an image, works in complete darkness, and can sense through non-metallic surfaces -- but it cannot tell who someone is. A camera with a vision model can read gesture, objects and appearance, but needs adequate light and creates a privacy obligation.

### Does a motion-tracking installation work in a dark venue?

Radar and depth sensors do, because they provide their own signal. RGB cameras do not without added illumination. If the venue is a dark launch environment or a night exhibition, rule cameras out early or budget for discreet infrared or visible fill lighting.

### Can a sensor be hidden inside the installation?

Radar can sit behind a non-metallic decorative skin and stay completely invisible, which is why it suits mechanical and sculptural pieces. Depth sensors and cameras need line of sight, though they can be recessed and shaded -- mounting geometry should be fixed during structural design, not afterwards.

### How accurate does sensing need to be?

Only as accurate as the interaction rewards. Sub-centimetre tracking is wasted on a wall where a guest waves an arm, and it costs calibration time. Define the smallest meaningful gesture in the design phase and specify the sensor against that, not against a spec sheet.

### Is facial recognition required for personalised interactions?

Usually not. Most personalisation runs on ephemeral session identity -- the system tracks that this is the same person for the next thirty seconds and then forgets. That delivers the experience without storing biometric data, and is far easier to clear with a brand's legal team.



## Related Pages

**See also:** [Motion & Body-Tracking Interactives](https://mlabart.com/solutions/motion-tracking-interactives/) | [AI Interactive Installations](https://mlabart.com/solutions/ai-interactive-installations/) | [Interactive LED Walls & Big Screens](https://mlabart.com/solutions/interactive-led-walls/) | [Mall & Pop-up Interactive Activations](https://mlabart.com/solutions/mall-popup-activations/)

---

**MLABART** -- Interaction - Imagination - Innovation

MLABART is a creative technology studio specializing in interactive installations, digital multimedia art, and immersive experiences. We take projects from creative direction to commercial delivery -- building audiovisual interactive installations, anamorphic 3D billboards (naked-eye 3D), kinetic mechanical installations, and AI-driven experiences for brand activations, exhibition spaces, and public environments.

- Web: https://mlabart.com
- Email: mark@touchworld-sh.com
- Phone / WhatsApp: +86 18917292695
- Canonical HTML version: https://mlabart.com/insights/motion-sensing-for-interactive-installations/