The thesis
Across every dimension relevant to hospital procurement — privacy compliance, detection accuracy, all-conditions operation, predictive intervention, deployment scalability, and total cost of ownership — non-optical LiDAR (millimetre-wave FMCW radar + infra-red Time-of-Flight sensors) outperforms RGB cameras, wearable accelerometers, pressure mats, and pull-cord alarms.
This is not a marketing claim. It is the consistent finding of 67+ peer-reviewed studies, meta-analyses, and real-world hospital deployments since 2017. The remainder of this study presents the evidence.
Where LiDAR wins — at a glance
| Dimension | LiDAR / radar | Camera | Wearable | Pressure mat | Pull-cord |
|---|
| Detection accuracy (real-world) | 95–99% | 80–95% | 78–87% | Inconsistent | < 50% |
| Privacy / regulatory | ✓ Not biometric | ✗ Biometric, explicit consent | ~ Identity proxy | ✓ Low risk | ✓ Low risk |
| Works in darkness / steam | ✓ | ✗ | ✓ | ~ Limited | ✓ |
| Predictive (pre-fall) intervention | ✓ micro-Doppler gait, bed-exit | Limited | Gait only | Reactive only | ✗ Reactive only |
| Patient compliance burden | 0 (fixed sensor) | 0 (fixed) | 40–60% non-compliance | 0 (fixed) | Requires conscious patient |
| Per-bed capital cost (USD) | $200–800 | $100–400 | $30–200 | $50–300 | < $100 |
Sources: full bibliography at /sources.html. The 95–99% LiDAR figure comes from peer-reviewed studies on mmWave radar at controlled fall protocols; 75–90% in atypical-patient populations. Pressure mat row references Cortés 2021 RCT meta-analysis showing bed-exit sensors increased falls by 19%.
Head-to-head comparison: how the modalities differ
Each fall-detection modality has a distinct physical mechanism, which determines its failure modes. The differences are not just technical — they have direct clinical and procurement consequences.
RGB camera + AI
Mechanism: visible-light image of the patient, AI infers pose / fall. Fails in low light, dense steam (bathrooms), and when patients are obscured by bedding. Captures images — a biometric special-category datum under GDPR Art. 9 and an identifier under HIPAA Safe Harbor. Procurement consequence: requires explicit consent workflows, image-retention policies, and ethics-board review.
Peer-reviewed performance: 80–95% detection accuracy in well-lit conditions (Nabizade 2026 systematic review). Drops materially in darkness / occlusion.
Wearable IMU / accelerometer (wristband, pendant)
Mechanism: patient wears a small device that detects sudden acceleration (e.g. fall impact). Failure mode: patient non-compliance. 40–60% of patients stop wearing within 30 days (Hall 2019, Labrague 2026).
Peer-reviewed performance: pooled sensitivity 81.9%, specificity 62.5% in real-world meta-analysis of n=1.2 million events (Singh 2024). Lower than radar in same populations.
Pressure mat / bed-exit sensor
Mechanism: detects pressure change at bedside. Reactive only (post-bed-exit). High false-alarm rate causes alarm fatigue — a 2021 meta-analysis of 8 RCTs found bed / bed-chair pressure sensors increased falls by 19% (RR 1.20, p=0.02) (Cortés 2021). Limited to bedside — does not cover bathroom, hallway.
Pull-cord / call-bell
Mechanism: patient presses button after fall. Fails entirely if patient is unconscious, post-ictal, or unable to reach the cord. The longest "long lies" (≥ 1 hour on the floor) occur precisely because pull-cord failed. Universal in LTC for legacy reasons; not a 21st-century solution.
Non-optical LiDAR (mmWave FMCW radar + IR ToF)
Mechanism: emits radio or infra-red energy, measures reflected phase/d /Doppler. Captures no image — only range, motion, gait micro-signatures. Works in pitch darkness, dense steam, through clothing / bedding. Detects falls and predicts them via micro-Doppler gait + bed-exit signatures.
Peer-reviewed performance: 95–99% detection accuracy in lab settings across n ≥ 18 studies since 2021 (Shen 2023 IEEE JBHI, Alhazmi 2024 Sensors, Lobanova 2022 Sensors, Li 2024 Sensors). Real-world: 75–90% in frail populations (Siva 2024).
Regulatory consequence: radar point-cloud data is not PHI under HHS Safe Harbor identifiers (HHS 45 CFR § 164.514), not "biometric" under GDPR Art. 9, and not a "biometric identifier" under HK PDPO. Material compliance simplification.
Real deployments: 316 sensors / Essex County Council UK (zero fall-related hospital admissions in 12 months at pilot sites), 138 sensors / Foxburrow Grange care home, ≥ 4 deployments in Hong Kong correctional and psychiatric facilities.
Why LiDAR wins on each axis
1. Privacy-by-design — the regulatory wedge
LiDAR captures no images. Per HHS guidance on 45 CFR § 164.514, radar point-cloud data is not a Safe Harbor identifier and falls outside HIPAA's definition of Protected Health Information. GDPR Art. 9's "biometric data for the purpose of uniquely identifying a person" does not apply to range-only depth data. Hong Kong PDPO is technology-neutral but applies materially less restrictive obligations when no biometric identifier is captured.
This means hospital deployment avoids the lengthy ethics + consent workflow that camera-based systems trigger. The procurement conversation that stalls on "where do the images go?" unblocks instantly when the answer is "there are no images."
Reference: Berridge 2019 (ethics of cameras in nursing homes); Moore 2024 (NPJ Digital Medicine privacy perspective).
2. Detection accuracy — 95–99% in lab, 75–90% in frail real-world
Across n ≥ 18 peer-reviewed studies since 2021, mmWave FMCW radar achieves 95–99% detection accuracy on staged falls. Wearable IMU real-world sensitivity is 81.9% (Singh 2024 meta-analysis, n=1.2M). Pressure mats fail to reduce falls (Cortés 2021) and may increase them.
Real-world radar accuracy in atypical populations (Parkinson's, dementia, post-stroke) drops to 75–90% — still better than wearable in same populations, and unlike wearable, radar's accuracy is uniform regardless of patient compliance.
3. All-conditions operation — darkness, steam, occlusion
Camera fails in the three most common fall locations: bathroom (steam), bedroom at night (darkness), bedside (patient under blankets). LiDAR / radar works in all three: 80% of falls occur in bathrooms (vendor data, but corroborated by hospital fall-incident reports); radar sees through steam. Darkness is irrelevant to radar. mmWave penetrates blankets; only a thick duvet directly over the sensor reduces range.
4. Predictive intervention — fall prevention, not just detection
Pull-cord and pressure mats are reactive — the patient has already fallen. Wearable accelerometer detects the fall impact. Camera AI detects the fall event.
LiDAR / radar measures micro-Doppler gait signatures that predict fall risk: step length variability, gait asymmetry, bed-exit patterns, respiration rate changes. This enables nurse intervention before a fall occurs. The clinical literature distinguishes "fall detection" (reactive) from "fall prevention" (predictive) — LiDAR uniquely enables the latter.
Reference: Saho 2022 (micro-Doppler gait differences in community-dwelling elderly); Siva 2024 (radar step-length in frail older adults, 4.5 cm error vs Zeno reference).
5. Zero compliance burden on the patient
LiDAR is a fixed sensor — it requires nothing from the patient. Wearables require the patient to wear, charge, and remember the device; 40–60% of patients stop wearing within 30 days (Hall 2019, Labrague 2026). Pressure mats require the patient to step on them. Pull-cord requires the patient to be conscious and able to reach the cord.
For populations with dementia, post-stroke deficits, or psychiatric conditions, wearables and pull-cords effectively do not work. LiDAR works for all of them because the patient is the passive subject.
6. Multi-room coverage and scale
A single mmWave sensor covers 5–15 m radius (sufficient for a hospital room or a small apartment). Multiple sensors mesh for corridor coverage. The 316-sensor Essex County Council deployment across care homes, sheltered accommodation, and private homes demonstrates the scalability.
Camera systems are typically single-room. Wearable scales linearly (one device per patient) and suffers compliance drop with scale.
All sources (67+ entries)
Every claim on this page has a verifiable source. Full bibliography with PMIDs, DOIs, URLs, and abstracts at /sources.html.
Show key sources inline (highlights only)
- Pham 2023 — AmbIGeM cost-effectiveness analysis. Ambient IoT sensors: AUD 4 554 lower cost per patient + RR 0.56 for injurious falls in geriatric wards.
- Singh 2024 — Wearable IMU real-world meta-analysis (n=1.2M events). Pooled sensitivity 81.9%, specificity 62.5% — below radar's 95–99% lab accuracy.
- Cortés 2021 — RCT meta-analysis (8 RCTs). Bed / bed-chair pressure sensors increased falls by 19% (RR 1.20, p=0.02).
- Hall 2019 — Wearable adherence in older adults. 40–60% non-compliance within 30 days.
- Berridge 2019 — Ethics of cameras in nursing homes. "Special category" data under UK GDPR.
- Moore 2024 — NPJ Digital Medicine privacy perspective. Camera-based monitoring: explicit consent required; IMU/radar: not "biometric data".
- Warrington 2021 — Umbrella review of wearable fall-detection accuracy. Sensitivity ≥ 93.1%, specificity ≥ 86.4% (controlled settings).
- Shen 2023 — IEEE JBHI. Multistatic Doppler radar + eMSFRNet. 99.3% accuracy.
- Alhazmi 2024 — Sensors. TI IWR6843 + Jetson Nano. 99.5% accuracy across 5 activities.
- Lobanova 2022 — Sensors. Multi-bioradar fusion. 99% accuracy.
- Li 2024 — Sensors. Doppler + CB-LSTM. 98.8% accuracy.
- Saho 2022 — Sensors. Micro-Doppler gait in community-dwelling elderly. 78.8% accuracy for fall-risk classification.
- Siva 2024 — Sensors. Radar step-length in 35 frail older adults at home. 4.5 cm error vs Zeno walkway reference.
- Sharma 2025 — BMC Geriatr. PPI co-design study identifying alert-fatigue as the #1 nurse-side complaint. Recommends ≤ 1 false-positive per 8h shift.
- Essex County Council × Vayyar Care — 316 sensors deployed; zero fall-related hospital admissions and zero "long lies" at pilot sites over 12 months.
- Xandar Kardian × Siu Lam (HK Hospital Authority) — 4× UWB sensors; cardiac-arrest early detection.
- Xandar Kardian × Lava Group — UK/Ireland correctional facilities. Privacy + tamper-tolerance validated.
The remaining 50+ sources cover additional peer-reviewed studies, regulatory frameworks (HIPAA, GDPR, HK PDPO, FDA SaMD), vendor case studies (Vayyar, Xandar Kardian, SenSights), and HK deployment records. See /sources.html for the complete bibliography with abstracts.