Speed Measurement Service

The Speed Measurement service enables automatic calculation of vehicle speeds from video recordings. For each tracked object, AITracker knows its precise location in every video frame (motion trajectory). By defining a measurement zone with known real-world dimensions, the system calculates how long each vehicle takes to traverse the zone and derives its average speed.

How It Works

Homography-Based Speed Calculation

AITracker uses homography transformation to convert pixel coordinates from the video into real-world distances. This geometric transformation maps a 4-point polygon in the video frame to known physical dimensions in meters.

The algorithm:

1. Polygon Definition - You define 4 points in the video that represent a known rectangular area on the road surface

2. Real-World Dimensions - You specify the width and length of this area in meters

3. Homography Calculation - The system computes a transformation matrix between image space and real-world space

4. Trajectory Tracking - As vehicles move through the polygon, their trajectory is tracked frame-by-frame

5. Speed Calculation - The distance covered (in meters) divided by time (in seconds) gives the speed

Entry and Exit Detection

The system detects when a vehicle enters and exits the measurement zone:

• Entry frame - First frame where the vehicle's bottom center is inside the polygon

• Exit frame - First frame where the vehicle leaves the polygon

The velocity is calculated as the mean speed between these two points, converted to km/h.

Configuration

Distance Zone Setup

To enable speed measurement, you must define a Distance object in your scene configuration with:

Parameter	Type	Description
trajectory	4 points	Four corner points of the measurement zone (relative coordinates 0.0-1.0)
width	float	Real-world width of the zone in meters
length	float	Real-world length of the zone in meters
relations_ids	list	IDs of lane relations to measure (connects to lane classifier)

4-Point Polygon Definition

The polygon must be defined as 4 corner points representing a quadrilateral on the road surface:

Point 1 ─────────── Point 2
   │                   │
   │   MEASUREMENT     │
   │      ZONE         │
   │                   │
Point 4 ─────────── Point 3

Guidelines for polygon placement:

• Place the polygon on a flat road section (no hills, curves, or elevation changes)

• Ensure the polygon covers the full lane width for the measured vehicles

• The longer the measurement zone, the more accurate the speed estimation

• Avoid areas with frequent vehicle occlusions

• Keep the camera angle as perpendicular to the road as possible for best accuracy

Real-World Dimensions

Accurate speed measurement depends on precise real-world dimensions:

• Width - The lateral dimension of the polygon (across the road), in meters

• Length - The longitudinal dimension (along the road direction), in meters

Tip: Measure the road surface dimensions in the field before setting up the video analysis. You can use road markings, lane widths, or known reference objects (car lengths, road signs) for calibration.

Technical Parameters

The velocity classifier has several configurable parameters:

Parameter	Default	Description
max_velocity	160 km/h	Velocities above this are discarded as invalid
min_time_in_the_zone_ms	500 ms	Minimum time vehicle must spend in zone for valid measurement
remove_frames_percent	10%	Boundary frames removed to reduce noise

Excluded Vehicle Categories

By default, the following categories are excluded from speed measurement:

• Bicycles and e-bikes

• Personal transporters (UTO)

• Scooters

• Pedestrians (all categories)

• Trains

• Warning vehicles

Measurement Accuracy

Aspect	Value
Typical accuracy	±5-10%
Best accuracy conditions	Flat road, perpendicular camera, clear weather
Processing speed	up to x5 (24h video in ~5h)

Important: This speed measurement is an estimation and is not certified for law enforcement or traffic violations. The values are suitable for traffic studies, planning, and general analysis.

Generated Output

Velocity.csv

The main output file contains:

Column	Description
Time	Timestamp when vehicle was measured
Direction	Movement direction (based on lane classification)
Label	Vehicle category
Velocity	Measured speed in km/h
Distance	Distance covered in meters
Timespan	Time spent in measurement zone (seconds)

Visualizations

AITracker generates:

• VelocityScatter.png - Scatter plot of all measured speeds over time

• Velocity.png - Average speed by hour

• VelocityPerCategory.png - Average speed by vehicle category

• VelocityDistribution.png - Speed distribution histogram

Report

The generated Report.docx includes:

• Speed statistics by direction

• Speed statistics by vehicle category

• Measurement precision analysis

• Speed limit exceedance summary (if limits configured)

Best Practices

1. Camera Placement

   • Mount camera high enough to see the full measurement zone

   • Avoid extreme angles that distort perspective

   • Ensure stable camera position (no vibration or movement)

2. Zone Design

   • Use minimum 10m length for reliable measurements

   • Match zone width to actual lane width

   • Position zone away from intersections or stopping areas

3. Calibration

   • Use known reference objects to verify dimensions

   • Test with vehicles of known speed if possible

   • Account for camera lens distortion

Limitations

• Speed accuracy depends on correct polygon dimensions

• Very fast vehicles (>160 km/h default) may be filtered out

• Curved roads reduce measurement accuracy

• Camera shake affects trajectory precision

• Not suitable for legal speed enforcement