# How a Fall Detection System Actually Works: Wearable Sensors vs. Radar Explained If you have ever wondered what happens in the seconds between a person falling and an alert arriving on someone's phone, you are not alone. Understanding how a fall detection system actually works -- not just what it promises -- helps families make more informed choices and set realistic expectations about what any technology can and cannot do. This article walks through the two main technical approaches to automatic fall detection: wearable accelerometer-based systems and ambient radar-based systems. We cover the underlying physics, the detection logic, and the practical accuracy implications of each approach. --- ## Why Understanding Your Fall Detection System Matters Falls are one of the leading causes of injury among older adults. According to the Public Health Agency of Canada, falls are the leading cause of injury-related hospitalizations among older adults in Canada. Beyond the physical injury itself, one of the most serious risks for someone living alone is the time that passes before anyone realizes they are on the floor. A fall that goes undiscovered for hours can turn a survivable injury into a life-threatening emergency. This is why automatic fall detection -- systems that identify a fall and trigger an alert without the fallen person having to do anything -- has become a meaningful category in senior safety technology. But the word "automatic" covers two very different technological architectures. Knowing how each one works helps you understand why they succeed in different situations and where each one's limitations lie. --- ## How a Wearable Fall Detection System Works The most widely deployed approach to automatic fall detection uses sensors embedded in a worn device -- a pendant, wristband, watch, or clip-on unit -- that the person carries or wears on their body. The core detection technology in most of these devices is an accelerometer, often working alongside a gyroscope. ### Accelerometers: Measuring the Physics of Motion An accelerometer is a microelectromechanical sensor that measures the acceleration forces acting on it along one or more axes. Most modern wearable fall detection devices use a three-axis accelerometer that simultaneously tracks forces along the X (left-right), Y (front-back), and Z (up-down) directions, expressed in units of gravitational force, or g. Under ordinary conditions, a person standing still registers approximately 1g on the vertical axis -- simply the pull of gravity. Walking produces rhythmic, oscillating acceleration patterns. Sitting down generates a moderate, controlled shift. A fall, by contrast, produces a characteristic and highly distinctive sequence of readings that most fall detection algorithms are specifically trained to recognize. ### The Fall Signature: Three Phases Rather than monitoring a single measurement, most wearable fall detection algorithms look for a three-phase pattern that together constitutes the fingerprint of a genuine fall event: 1. **Pre-impact free-fall phase.** When a person loses balance and becomes airborne, the accelerometer reading drops rapidly toward 0g across all axes. This near-weightlessness is a recognizable signature of a body in free fall. The duration is brief -- often well under a second -- but it is measurably distinct from any normal activity. 2. **Impact spike.** The moment the person contacts the floor, furniture, or another surface, the accelerometer registers an abrupt, high-magnitude acceleration spike. This can reach several multiples of normal gravity within a very short time window. 3. **Post-fall inactivity.** After a genuine fall, a person who is injured, disoriented, or unconscious typically remains still for an extended period. The algorithm cross-references the impact event with prolonged low-activity readings to distinguish a real fall from a device being dropped, a sharp bump against furniture, or another sudden movement. The specific threshold values -- exactly how low "near-weightlessness" must register, how sharp the impact peak must be, and how long inactivity must persist -- vary among manufacturers and are often refined through machine learning as systems accumulate more training data. ### Adding Gyroscopes: Rotation and Orientation Many systems supplement the accelerometer with a gyroscope, which measures angular velocity -- specifically, how quickly and in which direction the device is rotating. Gyroscope data helps the algorithm distinguish orientation changes and filter out ambiguous signals. A rapid rotation from an upright to a horizontal orientation, combined with the accelerometer signature described above, gives the algorithm considerably more confidence that a fall has occurred rather than a deliberate posture change such as lying down to rest. ### The Compliance Constraint The wearable fall detection approach has a fundamental limitation that no algorithm can fully resolve: the sensor must be on the person's body to detect anything at all. If the device is charging on the nightstand, sitting on a bathroom counter, or stowed in a pocket during a rushed morning routine, any fall that happens during that interval is completely invisible to the system. This is the compliance problem -- the practical gap between a device's technical detection capability and its real-world coverage, which depends entirely on whether the person is wearing it. Beyond compliance, wearable algorithms also face the challenge of false positives -- non-fall events that produce a sensor signature similar enough to trigger an alert. Dropping quickly into a chair, stumbling without falling, or vigorous physical activity can all create acceleration patterns that superficially resemble fall events. Managing false positives while maintaining sufficient sensitivity is an ongoing tension in the design of any wearable fall detection system. --- ## How a Radar-Based Fall Detection System Works A second category of fall detection system uses ambient sensing -- detecting activity throughout a room using radio waves rather than sensors attached to the person. The most capable ambient fall detection systems use millimeter-wave (mmWave) radar, typically operating at 60 GHz, combined with on-device artificial intelligence. ### The Physics of mmWave Radar Millimeter-wave radar works by transmitting short bursts of radio frequency energy and analyzing the signals that return after bouncing off objects and people in the surrounding space. The 60 GHz frequency band is well-suited to indoor fall detection: its wavelength of approximately five millimetres enables fine-grained spatial resolution while operating safely within regulated non-ionizing radio frequency limits used in many consumer and automotive applications. Two signal-processing techniques are particularly important for interpreting the returned signals: - **Doppler frequency shift** -- When reflected signals return from a moving object, their frequency is shifted slightly relative to the original transmitted signal. The direction and magnitude of this shift directly encodes the velocity of the reflecting surface. A person moving toward or away from the sensor creates a measurable Doppler signature, and the pattern of Doppler shifts over time provides detailed information about how the body is moving -- direction, speed, and changes in motion. - **Time-of-flight ranging** -- By measuring the tiny delay between the transmission and receipt of each pulse, the radar calculates distance to reflecting surfaces with millimetre-level precision. Combining measurements across multiple pulses and antenna elements enables a three-dimensional spatial representation of objects in the room, including a person's position, size, and approximate orientation. It is important to understand that mmWave radar at 60 GHz does not produce an image of the person. The raw data the sensor captures is a point cloud of radar reflections -- measurements of distances, velocities, and angles -- not photographs or video frames. ### On-Device AI and the Skeletal Model The raw radar point cloud is processed by an artificial intelligence model running directly on the sensor hardware, without transmitting raw data to external servers. This on-device processing enables real-time detection, and it means that only a processed event notification -- not a continuous stream of motion data -- ever leaves the device. This architecture is meaningful for privacy: the sensor does not record, store, or transmit any representation of what the person looks like. The AI model's task is to interpret the evolving point cloud as a representation of a human body's position and state of motion. Through training on large datasets of human movement patterns, the model learns to associate radar reflection patterns with distinct body postures and activities: standing, walking, sitting, lying down. More sophisticated implementations construct an abstract skeletal model -- a mathematical representation of estimated body joint positions inferred from the radar data -- that makes the person's posture and movement explicitly interpretable without capturing any visual likeness of the individual. ### Detecting a Fall with Radar AI The AI model monitors for the characteristic spatial and temporal signature of a fall event: a rapid decrease in the person's estimated height (the vertical component of the skeletal model collapsing quickly toward floor level), often accompanied by a high-velocity Doppler signature during the descent phase, followed by a persistent lying-down posture state. The model is trained to distinguish this pattern from deliberate lying down, which is gradual, controlled, and not accompanied by the abrupt high-velocity descent signature of a fall. Similarly, sitting down -- even quickly -- produces a very different spatial and velocity profile than a fall: the body descends only partway to the floor, the motion decelerates in a controlled way, and the resulting posture state is seated rather than prone. Because the sensor covers an entire room from its wall-mounted position, it monitors the full volume of the covered space -- regardless of where the person is within the sensor's field of view. Silvie, a Canadian wall-mounted radar sensor powered by a 60 GHz mmWave radar chipset, is one example of a commercially available system built on this architecture, with all motion processing handled on the device before any alert notification is generated. --- ## Key Differences Between Wearable and Radar Fall Detection Systems Understanding the technical differences between these two approaches helps clarify their distinct strengths and the situations where each performs most reliably. **Coverage model.** A wearable fall detection system monitors the body of the person wearing it -- the sensor data is body-relative, which means it travels wherever the person goes, indoors or outdoors. A radar-based fall detection system monitors a fixed physical space -- every point within the room, at all times, regardless of the person's behaviour. **The compliance variable.** Wearable fall detection depends on consistent use. A technically excellent device worn only part of the time has effective coverage only part of the time -- and no algorithm can compensate for a device sitting on the nightstand. Radar-based detection dramatically reduces the compliance variable: it operates continuously, regardless of whether the person is in bed, in the shower, or has simply forgotten about the system. **Privacy architecture.** Both approaches can be designed with strong privacy properties. Wearable devices collect body-movement data tied to the individual carrying the device. Radar-based systems collect spatial motion data from a fixed space, and modern on-device AI processes that data into an abstract skeletal representation -- no images, no audio, and no video of any kind. This makes radar-based sensors practical for spaces where cameras are socially and ethically inappropriate, including bathrooms and bedrooms, which are among the highest-risk areas for falls in the home. **Spatial scope.** Wearable systems can in principle detect falls anywhere the device travels -- on a walk, in a car, at a friend's home. Radar systems are room-specific by design. Each room requiring coverage needs its own dedicated sensor. This represents a meaningful difference in how families think about deployment. --- ## Practical Accuracy Implications of Each Approach A fall detection system's real-world accuracy depends on more than the strength of its detection algorithm. The performance gap between theoretical capability and practical outcomes is shaped significantly by the interaction between the technology and how it is actually used day to day. For wearable systems, the dominant real-world accuracy constraint is typically non-use rather than algorithmic failure. When worn consistently, modern wearable accelerometer algorithms perform well for sudden, high-impact falls. They are generally less reliable for slow or gradual falls -- situations where a person slides down a wall, sinks to the floor over several seconds, or experiences a fall with limited vertical free-fall phase -- because the near-weightlessness pre-impact signature is absent or significantly reduced. Gradual or slow-onset falls are a recognized challenge for wearable-based detection systems, which typically require a distinct impact threshold to trigger an alert. For radar-based systems, detection performance tends to be more consistent across different fall types, including gradual falls, because the algorithm observes the full three-dimensional spatial and velocity event rather than relying on body-mounted acceleration signatures. The practical constraint for radar is field-of-view coverage: furniture arrangement, room geometry, and sensor placement all affect how completely the system covers the monitored space. Proper sensor positioning for a given room layout is an important step in setup. Both approaches share the challenge of false positive management -- minimizing unnecessary alerts that erode trust and lead caregivers to mute or disable notifications over time. As of 2024, commercial systems have generally improved in this area as training datasets have grown more diverse and representative of real-world human movement patterns. --- ## What is the difference between manual and automatic fall detection? Manual fall detection requires the person who has fallen to take a deliberate action -- pressing a button on a pendant, for example -- before anyone is alerted. Automatic fall detection uses sensors to recognize the fall algorithmically and trigger an alert without any action from the person. The practical significance is substantial: someone who is unconscious, disoriented, seriously injured, or simply unable to reach a button still receives detection coverage under an automatic fall detection system, whereas a manual system provides no protection in those scenarios. ### Can any fall detection system detect every type of fall? No fall detection system -- of either type -- can reliably detect every fall in every situation, and reputable manufacturers are generally transparent about this. Wearable systems miss falls when the device is not worn and may be less sensitive to slow or gradual falls. Radar systems depend on sensor placement and room geometry for full spatial coverage. The honest framing is that any fall detection system meaningfully reduces undetected falls compared to no system at all, but no technology eliminates risk entirely. ### Do radar-based fall detection systems use cameras? No -- radar-based fall detection systems using mmWave technology do not use cameras or capture visual images of any kind. The sensor transmits and receives radio frequency signals; the data it produces is a set of range, velocity, and angle measurements from radio wave reflections, not photographs or video frames. On-device AI interprets this data to understand body position and movement without creating any visual representation of the person. This is one of the reasons radar-based systems can be deployed in bathrooms and bedrooms, where cameras are inappropriate. ### How does a fall detection system tell the difference between a fall and sitting down quickly? Wearable systems use multi-phase pattern recognition: the specific sequence of near-weightlessness, high-magnitude impact, and prolonged post-event inactivity is distinctive enough to filter out most non-fall events, though the sensitivity-versus-specificity balance varies between devices. Radar-based systems rely on spatial and velocity pattern recognition -- sitting down produces only a partial descent (to seated height, not floor level) with a controlled deceleration profile, whereas a fall produces a full-height rapid collapse with a high-velocity descent signature. Neither approach is perfect, but both have become considerably more accurate as AI training datasets have grown larger. ### Why do wearable fall detection devices sometimes generate false alarms? False alarms on wearable devices typically occur when a non-fall activity produces sensor readings that resemble the fall signature. Dropping heavily into a chair, stumbling without actually falling, vigorous exercise, or even rapid hand gestures can produce brief acceleration spikes that partially match the detection pattern. Manufacturers tune sensitivity thresholds to balance false alarm rates against detection sensitivity -- a system set to be highly sensitive to avoid missing falls will naturally produce more false alerts, and vice versa. Many systems allow sensitivity adjustment or use follow-up confirmation steps to reduce unnecessary alerts. ### Is fall detection for seniors effective in bathrooms? This depends on the type of fall detection for seniors being considered. Radar-based systems can be placed in bathrooms because they capture no visual information -- there is no privacy concern equivalent to a camera. Bathrooms are among the highest-risk areas for falls in the home, making coverage there particularly valuable. Wearable devices are technically capable of providing detection in a bathroom if the person is wearing them, but many seniors remove wearables before bathing or showering, creating a coverage gap at one of the most hazardous moments in the daily routine. ### Is mmWave radar safe for continuous use in the home? Millimeter-wave radar at 60 GHz falls within the non-ionizing radiation category -- it does not carry enough energy to damage biological tissue at the low power levels used in consumer fall detection sensors, in accordance with Health Canada's Safety Code 6. The technology operates within radio frequency exposure guidelines established by regulatory bodies such as Health Canada and the FCC, and it has been in broad commercial use in consumer electronics, automotive safety systems, and industrial sensing since at least 2017. Consumer fall detection sensors using 60 GHz radar are subject to the same national regulatory approvals as any radio-frequency device sold for home use. --- ## Understanding the Technology Supports Better Conversations A fall detection system is not a single technology -- it is a category that encompasses fundamentally different physical approaches to the same underlying problem. Accelerometer-based wearable systems and mmWave radar-based ambient systems each carry distinct technical strengths, accuracy profiles, and real-world constraints. Wearable systems travel with the person and work anywhere when worn consistently. Radar systems remove the compliance variable entirely, providing room-level coverage that operates regardless of user behaviour. Families considering fall detection for seniors benefit from understanding these distinctions before they encounter them through experience. The technology details covered here -- how the algorithms interpret motion, where each approach has coverage gaps, and how privacy is handled -- form the foundation for more productive conversations about which approach fits a given living situation. For deeper exploration of how ambient sensing compares against other monitoring approaches, or how radar systems handle specific room layouts, those topics are worth exploring in dedicated resources.