A motorcycle is not a small car. It carries a plate a fraction of the size, one or two riders who may or may not wear helmets and it goes where cars cannot: sidewalks, pedestrian zones, bus lanes, bike lanes. IREX treats two-wheelers as their own class, with their own detectors and their own rules.
The Class
Most traffic models learned two-wheelers as an afterthought, so a motorcycle in mixed traffic came out as a small car or as nothing at all. In release 4.36 IREX retrained CarTrack Pro to recognize two-wheelers as a class of their own, count the riders and detect whether helmets are worn. Release 4.37 extended that classification across the full set of tracking modes and 4.39 improved rider counting and helmet detection, including when a rider is only partially visible in the frame. The class is broad on purpose: any two- or three-wheeled motorized vehicle, including tricycles, scooters and mopeds.
The read is one pass. A first network finds the vehicle, a second reads its type and color and, for a motorcycle, the riders and their protective gear and a third and fourth find and read the plate. One read therefore returns the vehicle class, the plate, the rider count and the helmet status together, which is what a multi-rider or no-helmet rule is written against and what an investigator filters the archive on afterwards.
Bicycles are a separate problem and a separate module. BikeLaneTrack watches a bike lane for what should not be in it; the network behind it recognizes humans, bikes, cars, dogs and other object types and treats a bike with a rider close enough to it as one cyclist. Where the trained catalog stops, StreamVLM™ continues: a motorbike on the sidewalk and a two-rider robbery pattern are both prompts already running on our own home page.
Coverage
Motorcycles, scooters, mopeds and tricycles as their own vehicle class in every CarTrack Pro tracking mode, followed across the frame and, with Save tracks on, searchable by trajectory in the media player afterwards.
A motorcycle plate is a fraction of the area of a car plate and is often two-line or non-Latin. A regional plate model is selected per camera, a minimum plate size stops the platform guessing and since 4.39 a missing or unreadable plate raises its own event.
See License Plate Recognition →How many people are on the motorcycle, read on the same pass as the plate. More than one rider is the pattern a multi-rider rule is written against and the Motorcyclists filter returns those events on their own.
Whether each rider is wearing a protective helmet, recorded as an attribute of the event. Improved in 4.39 for riders only partly visible in the frame.
Motorcycles in car-only or bus-only lanes, in pedestrian zones, on crossings, or entering tunnels and highways closed to them, captured as a lane-restriction or forbidden-maneuver violation with the vehicle type as the rule.
BikeLaneTrack raises a Not a bike event for a pedestrian, a stroller or a skateboarder in the lane and a lane-restriction rule catches a car or a truck using it as a road.
On Camera
Two small motorbikes ride the sidewalk under an elevated railway in Bangkok while pedestrians press against the storefronts to let them pass. A car-shaped model misses this twice: it does not see the vehicle and it has no rule for a sidewalk. Here the rule can be a lane-restriction area drawn over the sidewalk with Motorcycles as the prohibited type, an ObjectTrack Pro Region entrance rule filtered to vehicles, or a sentence typed into StreamVLM™, which is the detector this frame was written for. Whichever fires, the event carries the frame, the plate where it is legible and the time.
The Rules
The class is the thing. Once a vehicle is read as a motorcycle, the platform’s existing enforcement machinery applies to it and the operator chooses which mechanism fits the road.
Draw a motion control area over the restricted lane, crossing or zone on a camera running CarTrack Pro with vehicle type recognition on, name the violation and select the vehicle types prohibited there. Written with Motorcycles as the type, it is a motorbike in a bus lane; written with every motor class, it is a car in the bike lane. The evidence package is the one every traffic violation gets.
Entry and exit lines at an intersection, with an optional vehicle-type filter, so a maneuver legal for a car can be a violation for a motorcycle, or the other way around. Exemptions by vehicle list or plate template let a permitted fleet through, which is how transit-priority rules already work.
Enable Count people on a motorcycle, detect safety gear in the CarTrack Pro settings, in any of the three tracking modes that support it. Every motorcycle read then carries its rider count and helmet status, searchable through the Motorcyclists filter, so more than one rider or no helmet becomes a query as well as an alert.
Camera
Two-wheeler detection is a pixels-per-meter question before it is a model question. The figures below are the documented CarTrack Pro and BikeLaneTrack ones.
Where It Runs
The same class of vehicle, a different problem in each deployment. In Peru the National Police traffic analytics deployment runs with two-wheeler detection because motorbikes are how the region moves and, sometimes, how it commits crime.
Helmet and rider-count compliance, motorbikes in restricted lanes and zones and bike-lane obstruction, each with an evidence package.
Read More →The two-rider pattern behind snatch theft and the plate read that turns a sighting into a route across the camera network.
Read More →A two-rider motorcycle alert arriving on the wall with the frame attached, while a patrol can still act on it.
Read More →Bike-lane monitoring and pedestrian-zone enforcement as one layer of the city operating picture.
Read More →Motorbikes and cars in the bus lane, on the same lane-restriction rule that catches a vehicle across a stop box.
Read More →Motorcycles as their own class at the barrier and in the occupancy count, on the same plate read as the cars.
Read More →Keep Reading
Two-wheeler analytics is a pixel-size question first. One camera on your busiest approach answers it faster than a specification sheet.
Yes and two-wheelers are their own vehicle class rather than a car that came out small. Motorcycle plates are a fraction of the area of a car plate, so the minimum plate size setting matters more here than anywhere: set the pixel width below which the platform should decline the read rather than guess. A regional plate model is selected per camera and two-line plates and non-Latin scripts such as Thai are supported. Where the plate is missing or unreadable, the platform says so with its own event rather than inventing characters.
Yes. Rider counting and helmet detection are part of the same CarTrack Pro read as the plate, enabled per camera with the Count people on a motorcycle, detect safety gear option in any of the three tracking modes that support it. Release 4.39 improved both, including when a rider is only partially visible. IREX publishes no accuracy figure for either; they are benchmarked on your cameras during the pilot.
Helmet detection is documented for motorcycle riders, through CarTrack Pro. For bicycles the documented module is BikeLaneTrack, which watches a lane for a cyclist and for what should not be there. A separate safety-gear check exists in ObjectTrack Pro for people on foot at work sites. If cyclist helmet compliance is the requirement, raise it before the pilot rather than assuming it.
The Motorcycles class covers any two- or three-wheeled motorized vehicle: motorcycles, tricycles, scooters and mopeds. Bicycles are handled separately by BikeLaneTrack and skateboards, strollers and pedestrians in a bike lane are reported as Not a bike.
Yes and there are two documented ways to write it. A lane-restriction rule with Motorcycles as the prohibited type on a motion control area drawn over the zone, or an ObjectTrack Pro Region entrance rule filtered to vehicles. Where the zone is irregular or the condition is more specific, such as a motorbike on the sidewalk, a StreamVLM™ prompt does the same job.
No. CarTrack Pro runs on standard CPUs through the proprietary Synet framework, though it is rated High in computational complexity, so IREX engineering sizes the servers against camera count and module mix. StreamVLM™ prompts, where you use them, need a GPU node.
That is the right question and the honest answer is that we measure it rather than quote it. IREX does not publish per-module accuracy figures, because a number from another city’s cameras tells you very little about yours. Accuracy is benchmarked on your own feeds during the pilot, against two or three success criteria agreed in writing before it starts and the measured numbers are what go into the contract.
Any camera that supports ONVIF or streams RTSP with H.264 or H.265 on a static IP can be connected, but connection is not the bar. Each module sets its own resolution, mount position, height and angle-of-view requirements in the camera requirements, the flagship modules need higher-resolution imagery of the scene than a bare stream and the recommended primary stream is 1920 × 1080 or 1920 × 1440. A site survey confirms every camera against the module it will run.
It runs on its own and alerts a person. Detections are signals for a human to verify: no response runs autonomously and the verification and decision are logged against the same Case ID as the detection.
