Bicycle detection at signalised intersections is one of the more technically awkward problems in urban traffic engineering. Cyclists generate a far weaker electromagnetic signature than cars, move more slowly than most detector algorithms expect, and often occupy parts of the lane that inductive loops weren't designed to cover. The result, in many older systems, is a cyclist who waits indefinitely because the controller doesn't know they're there.
That failure mode matters more now than it did a decade ago. Australian cities have invested heavily in cycling infrastructure, and transport authorities are under real pressure to make signalised intersections work for every road user, not just the ones driving a 1,500-kilogram vehicle. Getting bicycle detection right is a precondition for that.
Why standard vehicle loops miss cyclists
Inductive loop detectors work by measuring the drop in inductance that occurs when a metal object passes over a wire loop embedded in the pavement. A car produces a large, reliable signal. A bicycle produces a signal roughly 20 to 30 times weaker, depending on the bike's frame material, wheel size, and how close it passes over the loop's active region.
Loop geometry matters enormously. A standard 2-metre square loop cut into a single lane is optimised for vehicles straddling the centre. A cyclist riding close to the kerb, or positioned on a narrow shared path, may never pass over the high-sensitivity zone at all. Carbon fibre frames are particularly problematic: they contain almost no ferrous metal and can produce a signal indistinguishable from noise at standard sensitivity settings.
Sensitivity thresholds are set conservatively in most installations to avoid false triggers from passing trucks, adjacent traffic on parallel lanes, or embedded reinforcing steel in the pavement. Raising sensitivity to catch bicycles without generating constant false calls is possible, but it requires careful calibration against the specific site geometry and traffic composition.
Loop configurations designed for cyclists
The most straightforward fix is a dedicated bicycle detection loop cut to a geometry that maximises sensitivity for two-wheeled vehicles. Two formats are widely used in Australian practice.
The quadrupole loop uses a figure-eight or crossed winding pattern rather than a simple rectangle. It produces a characteristic double-peak signal as a bicycle crosses it, which the controller firmware can distinguish from vehicle noise. Quadrupole loops are better at detecting slow-moving or stationary bicycles, which is useful at stop lines where cyclists may be waiting at zero speed.
The narrow slot loop is simply a standard loop with a reduced width, typically 0.9 metres, positioned to sit under the bicycle lane's riding line rather than the vehicle lane centre. It trades some vehicle detection reliability for improved sensitivity to small objects. Where a dedicated bicycle lane already separates cyclists from general traffic, narrow slot loops are often the lowest-cost solution.
Video detection for cyclists
Video-based vehicle detection has been in operational use at Australian intersections for some years, but its performance on cyclists has improved substantially as processing hardware has grown cheaper and machine-learning classifiers have been trained on larger datasets of mixed traffic.
Modern video detection units can distinguish cyclists from pedestrians and vehicles using a combination of bounding box size, aspect ratio, speed, and trajectory. A cyclist approaching a stop line from a bicycle lane triggers a different classification than a pedestrian stepping off the kerb, and the controller can respond accordingly: extending green time for the relevant phase, or calling an additional bicycle phase where one is configured.
Video detection integrates naturally with vehicle detection technology in smart traffic systems already deployed at complex intersections, since the same camera feed can serve multiple detection purposes simultaneously. The trade-off is processing latency and the sensitivity of optical systems to lighting conditions. Rain, low sun angles, and night-time environments all degrade detection accuracy, and any installation relying solely on video must be tested across the full diurnal and seasonal range of conditions at that specific site.
Radar and microwave detection
Radar-based detectors use frequency-modulated continuous wave (FMCW) or Doppler radar to detect moving objects within a defined detection zone. They don't depend on pavement cuts, which makes them attractive for locations where loop installation is disruptive or impractical, such as bridge decks, bus lanes with high pavement stress, or intersections with frequent utility works.
Side-fire radar units mounted on signal poles can cover an approach lane from approximately 3 metres to 60 metres from the stop line, detecting both the presence and approach speed of cyclists. The key engineering challenge is occlusion: a large truck stopped at the line can mask a cyclist approaching behind it, and the radar returns from adjacent lanes can create false positives if the detection zone isn't correctly masked in software.
Some manufacturers pair radar with a dedicated bicycle sensitivity mode that lowers the minimum detectable radar cross-section. This mode needs to be disabled on approaches where pedestrians and cyclists share the same zone, since pedestrian detection at higher sensitivity can saturate the controller with calls.
Push-button and passive infrared options
Not every approach needs fully automatic detection. At low-volume intersections or locations where a dedicated bicycle phase is infrequent, a push-button actuator mounted at a height and position accessible to a cyclist without dismounting is a practical and compliant option under AS/NZS 2080. The button simply registers a bicycle phase call, and the controller services it according to the configured timing plan.
Passive infrared (PIR) sensors, more commonly associated with pedestrian detection, can also detect cyclists in slow-speed or stopped positions when mounted close to the stop line. They're less useful on high-speed approaches but work adequately in shared zones where cyclists are expected to slow before the detection point.
Connecting detection to signal timing
Detection hardware is only half the problem. A bicycle presence call is only useful if the controller's timing logic knows what to do with it. In practice, this means configuring a dedicated bicycle phase or an extended pedestrian phase, setting a minimum green time that allows a cyclist to clear the intersection at a realistic speed, and ensuring the phase is correctly represented in the signal group numbering scheme used by maintenance crews.
Where adaptive signal control is in use, bicycle detection feeds into the same demand inputs as vehicle detection. The adaptive algorithm needs to be parameterised to treat a bicycle call with appropriate weight: a single cyclist doesn't justify the same phase extension as a queue of 12 vehicles, but the cyclist shouldn't be starved of green time just because vehicle demand is high. Understanding how adaptive signal control responds to live traffic conditions is directly relevant to configuring these weights correctly.
Saturation flow modelling also needs adjustment where cyclists form a significant share of approach demand. A stop line that handles 1,800 passenger car equivalents per hour in a vehicle-only environment has a different effective capacity when cyclists are queuing and departing alongside general traffic. Getting those inputs right feeds directly into the timing plan design.
Commissioning and ongoing verification
Bicycle detection is one of the more frequently miscalibrated elements in a traffic signal system. A loop that was correctly set at commissioning can drift over time as pavement repairs alter its geometry, or as the sensitivity threshold is inadvertently reset during controller firmware updates. Video classifiers can degrade when camera positioning shifts after storm damage or pole work.
Operational verification should include direct testing with a representative bicycle and rider, not just a vehicle. Acceptance criteria need to specify the detection threshold explicitly: a system that detects a standard steel-frame bicycle 95% of the time but misses a carbon-frame bicycle entirely doesn't meet cyclist service requirements. Documenting these criteria in commissioning records and handover packages is as important as getting the hardware right on day one.
Transport authorities specifying bicycle detection should also consider how detection data is logged and audited. A controller that records bicycle calls alongside vehicle calls gives engineers real data on cyclist demand patterns, which informs future timing plan reviews and infrastructure investment decisions. Without that logging, the operational case for further cycling investment is harder to make.