2. Occupant classification and OOP detection

Out-of-position (OOP) suppression is only as good as the system's knowledge of who is sitting where. An airbag control unit that cannot distinguish a 5th-percentile female leaning forward from an empty seat, or a rear-facing infant seat from a 95th-percentile adult, will either deploy when it must not or suppress when it must protect. Occupant classification (OC) and OOP detection therefore form the sensing backbone of any FMVSS 208-compliant advanced restraint system. This chapter examines the four sensing families that dominate production architectures — weight (pressure) sensing, seat track position, belt buckle status, and capacitive/vision sensing — and how their outputs are fused into a deployment decision.

Why Multiple Sensing Modalities Are Required

No single sensor can satisfy the regulatory and functional requirements. A weight sensor alone cannot detect a child seat cinched tightly with a belt, because the belt tension transfers load into the seat structure and the measured mass may exceed the child-seat threshold. A belt buckle switch alone cannot tell whether the buckle is latched around a person or around a child seat. Capacitive sensing can detect the presence of a human body but is sensitive to moisture and seat heating elements. Vision sensing provides rich posture information but is affected by occlusion, lighting, and clothing. The industry answer is sensor fusion: each modality contributes a partial observation, and the classification algorithm combines them with confidence weighting.

Weight and Pressure Sensing

Weight sensing is the most widely deployed OC technology. Two implementation families exist:

• Bladder-based systems use a fluid- or air-filled bladder under the seat cushion connected to a pressure transducer. The measured pressure is proportional to the distributed load. These systems are simple and low-cost but have limited spatial resolution.
• Load-cell / strain-gauge systems place four sensors at the seat mounting points (two front, two rear). By summing the four readings, total occupant mass is estimated; by comparing front-to-rear and left-to-right ratios, the center of gravity (CG) location is computed. This CG information is what allows the system to distinguish a person sitting upright from a person leaning forward.

A typical load-cell OC system samples each channel at 10–50 Hz and applies a moving average to reject road-induced vibration. The classification thresholds are calibrated per seat and per vehicle. A representative (illustrative) calibration table might look like this:

These numbers are illustrative only; actual thresholds are determined by the OEM through sled testing and are stored in the airbag ECU calibration. The critical engineering challenge is drift: load cells exhibit thermal drift and creep, so the ECU must perform zero-point tracking when the seat is unoccupied (typically at ignition-off or after a defined idle period). If zero-point tracking is disabled or the seat is loaded during the tracking window, the classification will be biased.

Seat Track Position

Seat track position (STP) sensors report how far the seat is from its rearmost position. This is a direct OOP-relevant input because the airbag's deployment geometry is designed around a nominal seating position. A seat positioned far forward places the occupant closer to the airbag module, increasing the risk of inflation-induced injury. STP is typically measured by a Hall-effect sensor or a potentiometer integrated into the seat track, with a resolution of a few millimeters over a travel range of 200–300 mm.

STP is used in two ways. First, it gates the deployment: if the seat is within a defined forward zone, the airbag may be suppressed or deployed at reduced power. Second, it modifies the occupant classification: a forward seat position shifts the expected CG and changes the effective mass distribution seen by the load cells. The fusion algorithm must therefore treat STP as a context variable, not an independent decision.

Belt Buckle Status

The belt buckle switch is a simple normally-open or normally-closed contact that indicates whether the latch plate is inserted. Its role in OC is to confirm that the seat is occupied and belted. However, the buckle switch alone is insufficient: a child seat installed with the vehicle belt will show "buckled" even though the occupant is a child seat. The buckle signal is therefore used as a consistency check against the weight and capacitive sensors. If the buckle reports latched but the weight sensor reports empty, the system flags a fault and defaults to a safe state (typically suppression with a telltale warning).

Capacitive and Vision Sensing

Capacitive sensing measures the dielectric change caused by a human body in the seat. A conductive mat or electrode array in the seat cushion and backrest forms a capacitor with the occupant; the ECU measures the capacitance and derives a presence/proximity signal. Capacitive systems are excellent at detecting presence and can distinguish a human from a bag of groceries, but they are affected by moisture, seat heaters, and the occupant's clothing. They are often used as a complement to weight sensing, not a replacement.

Vision sensing uses a camera (typically mounted in the headliner or A-pillar) to capture the occupant's posture. A computer vision pipeline performs seat segmentation, head detection, and pose estimation, then classifies the occupant into categories such as "adult upright," "adult leaning forward," "child seat," or "empty." Vision provides the richest OOP information — it can detect a head within the airbag deployment zone — but it is computationally expensive and must meet strict latency and reliability requirements. Production vision-based OC systems typically run at 30 fps with a latency budget of under 100 ms from image capture to classification output.

Fusion Logic and Failure Modes

The fusion algorithm is where the engineering effort concentrates. A practical implementation follows a layered approach:

1. Signal conditioning: Filter each sensor, reject outliers, and perform zero-point tracking on the weight channels when the seat is confirmed empty.
2. Feature extraction: Compute total mass, CG longitudinal/lateral position, STP zone, buckle state, capacitive presence, and vision-derived posture class.
3. Classification: Apply a decision tree or a trained classifier (e.g., a small neural network or a set of threshold rules) to assign an occupant class with a confidence score.
4. Consistency and fault handling: Cross-check modalities. If the buckle says latched but weight says empty, or if vision says child seat but weight says adult, flag a fault and default to the safe state.
5. OOP determination: Combine the occupant class with real-time posture (from vision or CG dynamics) to decide whether the occupant is in the deployment zone.

Practical Verification Checklist

When validating an OC/OOP system on a vehicle, the following checks are representative of a production diagnostic routine. The specific values are illustrative and must be replaced by the OEM's calibration data.

The key takeaway is that occupant classification is a system problem, not a sensor problem. Each modality has blind spots; the fusion logic and the fault-handling strategy determine whether the restraint system behaves correctly across the full range of real-world occupants, child seats, and misuse conditions. The next chapter builds on this sensing foundation by examining how the classified occupant state is translated into deployment thresholds and suppression criteria.