Start with the metric, not the marketing claim

In busy lobbies, transit hubs, and employee entrances, speed gate tailgating detection is rarely a simple yes-or-no feature. Accuracy depends on how the gate senses movement, how the controller interprets edge cases, and how well the lane is tuned for real traffic rather than a demo environment. For a technical evaluator, the useful question is narrower: how often does the system correctly identify an unauthorized second person without generating enough nuisance alarms that operators start ignoring them?

That means your checklist should focus on measurable behavior under load. If a vendor only shows detection in clean single-file passage, you still do not know how the system will behave during shift changes, visitor surges, or bidirectional flow.

Check the sensing architecture before you look at software logic

The first filter is physical detection design. Ask what actually detects the second body: infrared beams, optical sensor arrays, overhead vision, side-mounted sensors, or a fused method. This matters because tailgating accuracy drops for different reasons depending on the sensing layer.

• Beam-based systems can work well in controlled lanes, but closely spaced users, bags, coats, or partial body overlap may confuse event timing.
• Vision-assisted systems may separate people better, yet lighting changes, shadows, reflections, and privacy constraints become part of the evaluation.
• Multi-sensor designs usually perform better in difficult traffic, but only if the fusion logic is visible and testable rather than presented as a black box.

If the vendor cannot clearly describe the sensor layout and detection zone boundaries, it is hard to trust any claimed accuracy figure.

Define what counts as a true tailgating event

A lot of evaluation mistakes happen here. One team treats “two people entering on one credential” as tailgating. Another also includes side-by-side attempts, reverse entry, hesitation in the lane, and a child moving close behind an adult. Detection performance will look very different depending on the event definition.

Before testing, lock down the event classes you care about: close-follow tailgating, piggybacking with physical contact, parallel entry, stopped-and-resumed motion, and wrong-way passage. Then require the vendor test log to map each alarm to one of those classes. Without that, “accuracy” is too soft to compare across systems.

Benchmark both detection rate and false-alarm behavior

A gate that catches nearly every tailgating attempt but alarms constantly during normal rush-hour use is not accurate in any operational sense. You need both sides of the picture.

Ask for test conditions with annotated video or event logs. A single aggregate number hides too much.

Stress the lane the way people actually use it

Real entrances are messy. People bunch up before the reader, pause to find badges, carry backpacks, roll suitcases, or turn sideways while talking. The most useful evaluation is a scenario test that deliberately introduces those conditions.

• High-density arrival waves at shift start
• Users with coats, carts, or large laptop bags
• Mixed visitor and employee traffic
• Two valid users walking with unusually small spacing
• One valid user followed by an unauthorized person matching the same pace

This is where many speed gate tailgating detection systems separate. A setup that is stable in static testing can become oversensitive once the entrance gets human.

Look closely at calibration and lane geometry

Detection accuracy is heavily shaped by installation quality. Lane width, pedestal spacing, sensor height, floor reflectivity, nearby glass, and approach angle all affect what the system sees. This is one reason benchmark results do not always travel well from one site to another.

Review the commissioning workflow. You want to know which thresholds are adjustable, who can change them, whether different presets exist for entry and exit, and how the system records tuning changes. If calibration is possible but undocumented, long-term performance drifts and nobody can explain why alarm rates changed after a maintenance visit.

Do not separate detection from access-control logic

The gate may detect motion correctly and still behave poorly if the credentialing logic, timer windows, or anti-passback rules are loose. Check how the system correlates a valid authorization event with a physical passage event. That linkage determines whether the alarm is meaningful, delayed, or misclassified.

Also confirm what happens after detection. Some environments need only an alarm to security operations. Others need barrier hold-close behavior, event video linking, or forwarding into a PSIM, VMS, or IBMS workflow. Accuracy is not only about sensing the event; it is about producing the right response with enough context to investigate it.

Use standards as a framework, not a shortcut

Relevant standards and interface requirements can help you assess integration maturity, safety, and interoperability, especially when the speed gate must sit inside a larger enterprise environment. But they do not, by themselves, prove tailgating detection quality. A system can integrate cleanly with ONVIF-adjacent video workflows or align with broader ISO and IEC-oriented procurement requirements and still perform poorly at lane-level detection if sensing and tuning are weak.

Use standards documentation to verify interfaces, event export, logging consistency, and maintenance discipline. Use field testing to verify actual detection accuracy.

What to do before you sign off

Run the evaluation in this order: define event classes, inspect the sensing design, test under peak-flow scenarios, measure false alarms alongside true detections, and review how calibration changes are controlled. If two systems look close on paper, the better choice is usually the one with cleaner logs, clearer tuning rules, and more predictable behavior in crowded lanes.

For technical evaluators, that is the practical standard. In high-traffic entrances, speed gate tailgating detection is accurate only when the sensing layer, logic layer, and site conditions have all been tested together, under pressure, with evidence you can audit later.