Good footage must show the full measurement geometry even when a bus, delivery car or truck is passing through the frame. Check the height, direction, obstacles and test frame before recording.
Quick selection
First, determine the type of measurement
Each road system requires a different frame. The three diagrams below show the starting point; detailed examples are provided below.
Cross-section
Typically 5–7 m. The camera looks along the road and the measurement area takes up most of the frame.
Crossroads
Typically 5-8 m. All required inlets, outlets and conflict areas must remain visible.
Roundabout
Typically 6–9 m. The frame covers the entire roundabout ring and all measured inlets and outlets.
Camera mounting height
Minimum installation height: 4 m
Typical working ranges:
5–7 m for the road section and smaller intersection
5–8 m for a larger intersection
6–9 m for roundabouts and extensive geometry
?Full geometryAll measured inlets, outlets and passage areas are visible.
?Large vehicleThe bus or truck does not cover the entire measurement area.
?No obstaclesSigns, poles and trees do not cross key trajectories.
?Effective framingThe road takes up most of the image; sky and elevations are limited.
Red example - bad installation: the measuring camera is clearly lower than the city camera. The difference in height increases the risk of being obscured by vehicles in the foreground.
01
Camera installation for a cross-section of the road
The camera should look along the road axis, cover the entire width of the measured cross-section and show the silhouette of the vehicle for several consecutive frames.
Green examplesExemplary or correct assembly.
Yellow examplesA difficult case; requires caution and a better point if available.
Red examplesInstallation unacceptable for the intended measurement.
How to assemble
Typical height: 5–7 m.
The road and vehicles take up most of the frame.
The camera looks obliquely along the movement path and covers the entire cross-section.
Trees, signs and vehicles in the foreground do not obscure the cross-section.
The final setting should be confirmed on a test motion recording.
Good cross-sectional examples
Perfect cross-section of an expressway or motorway: the camera is centrally positioned, looks along both roads and shows full silhouettes of vehicles in all lanes.
A model cross-section of an expressway: the central assembly point ensures clear geometry of both roadways, a full view of the lanes and a long section of vehicle observation.
Exemplary editing: the road dominates the frame, the vehicles are large and remain visible for a long distance.
Model highway cross-section: the camera looks along the road, covers both directions and provides a long, uninterrupted observation period.
Model cross-section: the vehicle occupies a large part of the frame, maintains a clear silhouette and remains visible on a long section of the road.
A good cross-section of the highway: the camera looks along the road, and the long visible section allows you to observe vehicles in both directions.
The high mounting point provides a full view of both directions of traffic and the entire cross-section of the road.
The camera looks along the road and the vehicle remains large and legible in subsequent frames.
A full cross-section and an appropriately long observation period.
Vehicles remain visible in front of and behind the measurement line.
A model cross-section of a railway crossing: you can see the entire width of the road through both tracks, barriers and vehicles waiting in front of the crossing. The railway infrastructure does not cover the measured traffic path.
Correct cross-section at the railway crossing: both traffic directions and the full width of the road remain visible, and vehicles and cyclists have clear silhouettes.
The cross-section is legible and suitable for counting.Note: A point closer to the center of the cross-section and a straight line along the road axis would be even better.
Red examples: ineffective cross-section
UnacceptableToo far from the road and too much unnecessary background
The road occupies too small a part of the image
AITracker can recognize vehicles, but they are small and the camera looks at the highway from a long distance and at an unfavorable angle. Much of the resolution and processing power is used on the sky, grasses and surroundings instead of the road.
How to improve: place the camera closer to the highway, point it more along the road and limit the frame to the measured cross-section.
UnacceptableToo low and across the track
Camera too low and perpendicular to the road
The vehicle stays in the useful part of the frame for a very short time, and the closer vehicle may obscure the further lane. Much of the image includes an insignificant shoulder rather than a longer stretch of road.
How to improve: raise the camera and point it diagonally along the road axis so that the vehicles remain visible for several consecutive frames.
UnacceptableLow transverse view of a multi-lane roadway
Across multiple lanes from a low point
Vehicles in the closer lanes can obscure traffic in the far lanes, and each car stays in the frame for a very short time. The sidewalk, bike path and lawn unnecessarily take up a large part of the image.
How to improve: find a higher point closer to the road and observe the stripes diagonally along the direction of travel.
02
Camera installation for the intersection
The frame must include all measured entrances, exits and the area where vehicles change direction. Just seeing the entrance lines is not enough.
How to assemble
Typical height: 5–8 m.
A point outside the axis of signs, poles, trees and signaling elements.
The frame includes all measured inlets, outlets and the center of the intersection.
A tall vehicle close to the camera does not cut off the opposite side.
A large intersection may require several cameras.
Good examples of intersection
Exemplary intersection assembly: a high point provides a clear view of the center of the intersection, measured entrances and traffic lanes. The vehicles have large, distinct silhouettes even on wet surfaces.
A sample installation of a smaller intersection: the camera covers the full width of the main road, the side entrance and the junction area. The high point reduces the risk of being obstructed by vehicles.
Model assembly of an intersection with crossings: the high point includes roadways, islands, the full width of crossings and pedestrian access zones. One frame allows you to analyze vehicle and pedestrian traffic.
A model assembly of a wide intersection: both side entrances, passages, all lanes of the main road and a long access section are visible. The high point maintains connections even when heavy vehicles pass by.
Correct height and frame including inlets, outlets and the course of the relationship.
Exemplary assembly: central view, full geometry and large image reserve around the intersection.
The same reference point without any relationships. Tall vehicles do not cover the entire opposite side.
Good editing: the four measured arms, the center of the intersection and the course of the relationship remain visible in one frame.
Yellow examples of intersections
A borderline example: the intersection is still legible, but the perspective is less favorable and the frame space is small.
The camera is too far from the intersection: the vehicles in the coverage area are small and the bus in the foreground obscures part of the entrance.How to improve: place the camera close to the intersection, on a pole, at a height of 5-7 m.
Bad intersection examples
Negative examplePermanent obstruction in the axis of observation
Estimated height: 2.5–3.5 m
The sign covers the inlet
The STOP sign disc is located in the observation axis and covers part of the measured inlet and the course of the trajectory. Much of the frame includes the sky and buildings instead of the road.
How to improve: raise or move the camera off-axis of the sign.
Negative exampleFixed obstacles block the inlet
Estimated height: 3.8–4.5 m
The sign covers the inlet
The sign and the advertising column are located in the observation axis. They cover part of the inlet and vehicles approaching the intersection, which creates permanent blind spots.
How to improve: raise the camera or move it off the axis of signs and advertising poles.
03
Camera installation for a roundabout
The roundabout should be observed as one system: a ring and all measured inlets and outlets. A single cross-section view does not reliably reconstruct the relationship.
How to assemble
Typical height: 6–9 m.
The entire ring and measured arms are inside the frame.
Monuments, islands, signs and vegetation shall not obstruct opposing inlets.
For large geometries, several cameras should be used.
Good examples of roundabouts
Exemplary roundabout installation: a high point provides a full view of the ring, entrances, exits and traffic lanes. The low central island does not obscure vehicles on the opposite side.
Exemplary roundabout assembly: the entire ring, entrances, exits and pedestrian and bicycle crossings remain visible. The low island does not cover the traffic paths, and AITracker maintains the continuity of the relationship.
Good installation: a high point covers the ring and arms of the roundabout, and vehicles remain visible for a long part of the relationship. The frame is correct, although the wide lens introduces noticeable distortion.
Yellow examples: difficult roundabouts
WarningVegetation covers part of the ring
Bushes in the middle of the roundabout
AITracker can connect part of the trajectory after a short occlusion, but this is a difficult case and increases the risk of interruption or incorrect connection of the trace.
How to improve: set the camera higher and at such an angle that you can see the driving path above the vegetation.
WarningThe tree creates a permanent blind spot
The tree on the central island
The tree trunk and crown cover part of the ring. The object disappears in a fixed place, which makes it difficult to maintain the continuity of the trajectory.
How to improve: move the observation point outside the tree axis or use a second camera.
WarningObservation point too low and distant
Bushes cross the track
AITracker's algorithms can connect the tracker after the vehicle comes out from behind the bushes, but installation remains risky and provides little headroom for denser traffic.
How to improve: mount the camera closer to the roundabout and higher to observe the ring more from above.
WarningThe opposite side of the roundabout is at the limit of readability
Estimated height: approximately 5–7 m
The opposite side of the roundabout is on the border
The height is sufficient, but the opposite part of the roundabout and the further entrance are small and close to the edge of the image. The material can be useful, but provides little margin in denser traffic and occlusions.
How to improve: rotate the camera slightly to the right or change the mounting point so that the entire roundabout is closer to the center of the frame.
Red examples: assembly unacceptable
Passenger cars: the roundabout is visibleBus: almost the entire roundabout disappearsUnacceptable
Low mounting only works for passenger cars
For small vehicles, the roundabout geometry looks clear. However, a bus or other tall vehicle passing close to the camera obscures the ring, further inlets and configured relationships. The material is not suitable for reliably measuring full relationships for all vehicle categories.
How to improve: mount the camera higher so that it looks above the bus, or divide the roundabout observation between several cameras.
The van covers the roundabout and the intersectionsSame error: the entire entrance disappears from the imageUnacceptable
Estimated height: 2.8–3.4 m
Camera too low: a large vehicle is blocking the roundabout
The main cause of the error is the mounting point being too low. A tall vehicle passing close to the camera covers the center of the roundabout, the opposite road, the configured relations or the entire entrance.
How to improve: mount the camera higher to look over vans and trucks, or use multiple cameras.
04
Speed measurement
The speed is determined over a defined section of the image. The camera must be stationary and the beginning and end of the section must correspond to points on the same surface.
How to define the measured road section
Use two easily visible ground points and measure the actual distance between them.
Drive the section along the track, not diagonally between the lanes.
Both ends must be on the same plane of the road.
The vehicle should be visible continuously throughout the entire section.
Do not change the camera position or zoom after calibration.
Good examples of measured road sections
Correctly defined speed measurement section consistent with the vehicle's driving path.
Correctly designated speed measurement sections for both directions of traffic.
Correct measuring section on a city road: the beginning and the end lie on one plane and are consistent with the driving path.
05
ALPR combined with traffic measurement
A stationary camera observes passing vehicles. The same material can be used for vehicle classification and license plate recognition if it shows the full silhouette of the vehicle and a sufficiently large, sharp license plate at a small angle.
How to set up a still camera
Point the camera along the road so that the vehicle and license plate remain visible for several consecutive frames.
Maintain the full vehicle silhouette needed for traffic classification.
Keep both the horizontal and vertical viewing angle to the license plate as small as possible.
Set a short exposure time appropriate to the speed of the vehicles.
Evaluate the test on individual frames: the number must be human-readable.
Reference traffic counting and ALPR from one camera
One camera captures the full silhouette of the vehicle for traffic classification and the rear plate at a low angle for ALPR reading. The long view along the road provides several consecutive frames for both analyses. The license plate in the published copy has been anonymized.
An exemplary view for traffic in both directions: the camera observes a long section of the road, preserves the full silhouettes of vehicles and provides a plate reading zone before reaching the edge of the frame.
Readability check
The license plate must be readable in a single frame
In stationary traffic measurement with ALPR, a visible vehicle is not enough. The camera must register the characters on the board without blurring, overexposure and strong perspective deformation.
The most important rule
If the array number is not human-readable in any frame, the AI has no reliable data from which to reconstruct it. Blurred text should not be guessed.
A place for a good frame of the boardAn example of a sharp, readable plate of a vehicle passing in front of a stationary camera will be useful.
06
Measurement of parked vehicles with a camera behind the glass
The camera is mounted inside a measuring car that drives along rows of parked vehicles. The frame should be limited to the boards, the reflections of the glass should be limited and the driving should be carried out slowly and smoothly.
How to install a camera behind the glass
Mount the camera steadily, close to clean glass.
Point the lens towards the row of parked vehicles, as close to the signs as possible.
Reduce reflections from the dashboard and interior elements.
Narrow the frame so that the license plates are large enough.
Drive slowly and check test frames before starting the actual measurement.
Good examples of drive-by surveys of parked vehicles
An exemplary tour of a parking lot: the camera works behind the windshield of the car, and the drive leads along rows of parked vehicles. The arrays are located close to the observation axis and remain visible for subsequent frames.
Correct frame of the mobile ALPR: the camera is stable behind a clean window, and the measuring car passes close to parked cars. The license plates are large enough and viewed from a slight angle.
A camera mounted inside the vehicle records parked cars during the detour.
The camera is aimed at a row of parked vehicles and records their license plates as they pass by.
Incorrect recordings through the glass at night
Too wide frame. The license plates are too small and blurry; the signs cannot be reliably read from this distance.
Movement and reflections. The movement of the car and reflections in the glass cause blur and x-rays; there is no single readable frame of the board.
Not enough details. The license plate occupies too small a part of the image, so individual characters cannot be reliably read.
How to prepare a reliable recording of parked vehicles?
Stabilize the camera close to clean glass.Lock image orientation, remove dashboard reflections, and don't position the lens behind a tinted or highly reflective piece of glass.
Reduce exposure time.It's the shutter speed, not the high FPS itself, that limits motion blur. A shorter exposure requires more light.
Illuminate the license plates from the outside.Place the IR illuminator close to the lens axis and point it at the reading area. The camera must support infrared; IR left inside may reflect off the glass.
Narrow the frame and reduce the viewing angle.The board should occupy a sufficiently large part of the image and the camera should look at it as little obliquely as possible.
Slow down the crossing.Slower speed reduces blur and increases the number of readable frames.
Consider two cameras.The first one registers plates, the second one records entire vehicles and the surroundings. Materials can be related by time and location.
Illustrative diagram. The specific set and settings must be confirmed in a short overnight test before the actual measurement.
07
Night measurement
A night shot is correct when the silhouette of the vehicle and its features are still visible. Headlights or a bright spot alone are not sufficient for classification.
Night measurement condition
The measurement area must be illuminated with infrared (IR) and the camera must record an infrared image.
Good nighttime examples
A good nighttime frame: the full silhouette of the vehicle remains legible despite the lights, and the vehicle occupies a sufficiently large part of the image.
Good nighttime road cross-section: vehicles are visible for a long distance, and the lighting preserves the lines and silhouette.
A good nighttime frame of a truck: the full silhouette and body features remain recognizable despite wet roads and lights.
A good example at night with IR illumination: despite bright headlights, vehicle silhouettes, lanes and road geometry remain visible.
A good example at night: despite the bright rear lights, you can still see the shape of the moving vehicle, its position on the road and the course of the observed road.
Warning examples: borderline recognizable
The frame is at the limit of recognizability. High speed and high blur.
The frame is at the limit of recognizability. High speed and high blur.
The frame is at the limit of recognizability. High speed and high blur.
The vehicle category is recognizable in the near lane, but the far lane does not provide an equally reliable image.
The measurement is useful in the near lane, but in the far lane the vehicle silhouette is blurred and dominated by the lights. Category recognition in the further lane is less reliable.
Warning example: The vehicle is visible, but the movement causes strong blur. The category may be difficult to reliably define.
Bad nighttime examples
The vehicle as a ball of light will not be recognized!
Only the lamps visible are not enough for the AI!
Bad example: mainly points of light are visible and the silhouettes of vehicles disappear. Such material does not allow for reliable category recognition.
08
Weather conditions
Rain, fog, snow, low sun and dirty lenses can turn a good daytime frame into material unsuitable for analysis.
What to check
The camera housing must not introduce droplets or reflections in front of the lens.
The camera should have a stable, wind-resistant mounting.
Low sun and overexposure do not disqualify the frame if the vehicle silhouettes and road geometry remain legible.
After rainfall, check the sharpness, contrast and cleanliness of the glass or lens.
A good example in the rain
Rain and wet surfaces do not exclude measurement. Vehicle silhouettes, lane markings and the entire observation area remain legible.
Low sun: correct frame despite overexposure
A good example: despite low sun, overexposure and flare, the silhouette of the vehicle, its position and the geometry of the road remain legible.
Snow at night: a difficult but quantifiable case
Snowfall, wet surfaces, drops in front of the lens and strong reflections from the headlights clearly deteriorate the image. In this material, AITracker still maintains trajectories and recognizes vehicle categories, but such measurement requires careful control of the results.
The heavy vehicle maintains a clear silhouette and trajectory despite snow and drops in front of the lens.
The algorithm maintains the driving path when the road surface is wet, there are reflections and the image is partially obscured.
The category remains recognizable despite snow and direct headlight glare.
The passenger car is detected and tracked through the measurement area despite difficult weather conditions.
Go from knowledge to your own measurement
Download AITracker software that automatically recognizes vehicle categories and counts them from video recordings. A sample video demo with results built into the program will allow you to learn about the program for free.
See for yourself the quality of AITracker and test the program!
UnacceptableToo far from the road and too much unnecessary background
