How to Calculate Vehicle Speed From Video: A Step-by-Step Guide

To calculate vehicle speed from video, you need two measurements: the real-world distance a vehicle travels and the time it takes to cover that distance. Divide distance by time, then convert the result into km/h or mph. A phone camera and the AI Traffic Count app can automate the difficult part by detecting vehicles, tracking them between two virtual measurement gates, and calculating an estimated speed.

This guide explains the complete workflow—from choosing a camera position to checking the results—and shows how to avoid the setup mistakes that have the greatest effect on accuracy. You can use it for traffic observations, before-and-after studies, neighborhood road-safety projects, driveway monitoring, or general traffic research.

The basic calculation is simple:

Speed = distance ÷ time

If distance is measured in meters and time in seconds, the result is meters per second. Use either of these conversions:

• Speed in km/h = distance in meters ÷ time in seconds × 3.6
• Speed in mph = distance in meters ÷ time in seconds × 2.23694

For example, a vehicle that travels 10 meters in 1.25 seconds is moving at 8 meters per second. That equals 28.8 km/h or approximately 17.9 mph.

The formula is not the difficult part. Reliable results depend on identifying the same point on the vehicle at both measurement gates, measuring the real road distance correctly, and determining the elapsed time from the video frames. AI-based tracking makes those timing and detection tasks much faster.

A video is a sequence of frames with known timing. A speed-estimation system detects a vehicle in one frame, associates it with the same vehicle in later frames, and determines when it crosses each measurement gate. The time between those crossings becomes the time part of the speed formula; your measured gate separation supplies the distance.

This approach combines object detection, multi-object tracking, frame timing, and a conversion between image coordinates and the real road. Technical implementations may use a real-world scale such as meters per pixel or calibrated road coordinates. Together, these measurements allow the system to translate movement across video frames into an estimated real-world speed.

Research into monocular-camera speed measurement emphasizes that converting positions in a two-dimensional image into real-world coordinates is one of the central challenges. A practical alternative uses two virtual lines with a measured real distance between them—the same basic distance-over-time principle used in this guide. See the open-access Scientific Reports study on monocular vehicle speed measurement for a more technical explanation.

No ordinary camera setup should be assumed to produce perfect measurements. Treat the result as an estimate and improve repeatability with a controlled setup.

• Stabilize the camera. Even small movements change the apparent position of the gates and vehicles.
• Measure on the road plane. The entered distance must follow the vehicle’s actual route between the gates.
• Use a longer practical distance. Very short distances magnify small timing and placement errors.
• Keep both gates in a similar depth range. Extreme perspective makes image-to-road conversion more sensitive.
• Use adequate resolution and frame rate. More clear frames around each crossing give the tracker more information.
• Avoid occlusion. Do not let parked vehicles, vegetation, poles, or other traffic hide the measurement zone.
• Test with a known reference. If possible, record a vehicle traveling at a verified steady speed and compare the estimate.
• Repeat the measurement. Look for consistent patterns across several passes and investigate obvious outliers.

Resolution, frame rate, tracking quality, environmental conditions, perspective, and camera specifications can all influence the result. Validation against known reference data is the best way to understand the margin of error for your particular setup.

You can calculate speed manually by choosing two visible road points, counting the frames between crossings, dividing the frame count by the video frame rate, and applying the distance-over-time formula. This can work for one vehicle, but it becomes slow and inconsistent when many vehicles must be reviewed.

AI-assisted analysis automates vehicle detection, tracking, crossing-time measurement, and repeated calculations. That makes it more practical for traffic observations where the goal is to understand patterns across many vehicles. It can also preserve individual detections so you can review questionable results instead of relying only on an overall average.

Video-based results are speed estimates, not automatically certified enforcement measurements. Do not use them to accuse an identifiable driver, issue penalties, or claim legal proof unless your local authority and an appropriately qualified expert confirm that the complete method, equipment, calibration, and evidence handling meet the applicable requirements.

Respect privacy and recording laws. Requirements differ by country and location, especially when filming public roads, neighboring property, people, or license plates. Record only where you are permitted to do so, minimize unnecessary personal data, and share aggregated traffic findings whenever individual evidence is not needed.

Not reliably. The video should have stable timing, a stationary viewpoint, a visible road section, and enough information to determine a real-world distance. A moving camera, edited playback speed, missing frames, severe perspective, or unknown frame timing can make a useful calculation impossible.

There is no single ideal distance for every road. Use the longest practical distance over which vehicles remain clearly visible and normally maintain their lane. Avoid distances so short that a one-frame timing difference would substantially change the result.

A phone camera can provide useful estimated speeds for observation, screening, and traffic studies, but it should not automatically be treated as a replacement for a calibrated or legally approved radar, lidar, or enforcement system. The appropriate tool depends on the accuracy and evidentiary requirements of the project.

Potentially, if the camera was stationary, the original frame timing is intact, the vehicle crosses two visible reference points, and you can measure the real distance between them. Compressed, variable-frame-rate, time-lapse, or exported footage may require extra verification.

AI Traffic Count turns an iPhone or iPad into a practical traffic-analysis tool. Use it to count vehicles, review detections, and estimate speeds from a carefully measured video setup—without installing permanent roadside hardware.

For a broader introduction, read how the app automatically counts vehicles in video. If you are evaluating a road-safety intervention, you can also learn how to track the effect of traffic-calming measures.