Examinations

How is vehicle speed calculated from dashcam or surveillance video?

A speed taken from video is the distance a vehicle covered divided by the time between the frames that show it. Both numbers carry error. The time has to come from the frame timing stored in the original file, because many recorders do not capture frames at even intervals, and the distance has to come from measurements of the scene. US courts have allowed such calculations and have barred them when the inputs were guesses.

A dashcam on a windshield frames a blurred evening intersection

How does a reconstructionist calculate speed from a video?

The reconstructionist fixes the vehicle's position in two frames, measures the distance between those positions and divides it by the time between the frames. The result is an average speed over that stretch of road.

Gray Beauchamp and five co-authors at Kineticorp set out the method in Speed Analysis from Video: A Method for Determining a Range in the Calculations (SAE Technical Paper 2021-01-0887). Average speed equals distance divided by elapsed time. The vehicle's positions come from camera-matching photogrammetry: a three-dimensional model of the scene is aligned with each frame, and a model of the vehicle is fitted into the frame. For a camera that records at a constant rate, the elapsed time is the frame count divided by the frame rate. In their tests the authors assigned a time to each frame from a timing light they built and placed in view of the cameras, and for the variable-rate camera they also used frame-to-frame timing extracted with iNPUT-ACE software.

The Scientific Working Group on Digital Evidence (SWGDE) warns against the shortcut of dividing a total distance by the running time of the clip. Section 4 of its 2020 document Considerations for the Use of Time-Based Analysis of Digital Video for Court says a speed computed that way may not match the movement captured in the video, because the number, duration and timing of individual frames vary.

The number of frames between the two positions changes the result. In a hypothetical in the Kineticorp paper, positions one frame apart gave a range of approximately 12 to 88 mph, plus or minus 76 percent. Positions ten frames apart gave 46 to 55 mph, plus or minus 7 percent.

The speed is an average over the measured stretch. A vehicle that brakes inside that stretch was moving faster at its start and slower at its end. The frames chosen decide which speed the report gives.

Why is the frame rate in the file properties not enough?

Many recorders store frames at uneven intervals, and players, exports and screen captures change the timing. The time between two frames is read from the timing data of the original file, frame by frame.

SWGDE's Core Technical Concepts for Time-Based Analysis of Digital Video Files (version 1.0, September 17, 2020) defines frame rate as "a measure of the video display rate in frames per second" and notes that it is typically expressed as an average. Section 5.2 states that streams with variable frame rates include frames of differing durations. The presentation time of each frame is calculated from the durations of the frames before it, and the document names a frame information report, such as one generated with FFmpeg, as a tool for reading those values.

The companion document for court use adds two warnings. Under section 5.6, devices may record "an unequal number of images during equal intervals of time" even when the user settings specify a rate. Under section 5.3, players differ in what they display, and even the manufacturer's own player may not play the video back at the correct speed.

The Kineticorp tests measured the effect. A Google Pixel II phone recorded at 29.89 to 30.12 frames per second, and a Sony RX100 II camera held a constant 29.97. A Vivint home security camera averaged between 1.81 and 5.53 frames per second across the test videos. At 11 mph, speeds from the phone and the Sony camera agreed with the VBOX GPS reference in all segments. For the Vivint camera at 21 and 26 mph, some calculated ranges fell entirely outside the measured speed, even with per-frame timing extracted by iNPUT-ACE software.

The same issue came up in Allcorn v. Beach, a 2025 federal case in Mississippi. The plaintiff's reconstructionist described the tractor-trailer's dash cam footage as approximately 10 Hertz, and his initial report noted that the footage did not have evenly spaced time.

Which copy of the video does the analysis need?

The native file exported from the recorder, with its metadata. A phone recording of a monitor or a re-encoded clip can carry a different frame rate and missing frames.

Section 5.1 of the SWGDE court document says the source of the digital video "should be known and verified," and that without the original file the effects of conversion may not be apparent. Under section 5.4, transcoding changes the metadata stored in the file container and may change the frame rate while the pictures look the same. Under section 5.5, transcoding, screen capture and data transfer may cause frame loss, which "will affect calculations and timeline sequence reconstruction."

In federal civil cases the request goes out under Rule 34 of the Federal Rules of Civil Procedure. Rule 34(a)(1)(A) covers electronically stored information, including images and sound recordings, and Rule 34(b)(1)(C) lets the requesting party specify the form in which it is produced. If the request names no form, Rule 34(b)(2)(E)(ii) has the responding party produce it in the form in which it is ordinarily maintained or in a reasonably usable form. A request that names the native export with metadata closes that gap.

The Federal Rules of Evidence treat a copy as a duplicate when it is produced by a process that accurately reproduces the original, Rule 1001(e). Under Rule 1003, a duplicate is admissible to the same extent as the original unless a genuine question is raised about the original's authenticity or the circumstances make admitting the duplicate unfair. For a speed calculation the question is narrower: whether the copy kept the frame timing of the original.

Recordings get lost. Rule 37(e) applies when electronically stored information that should have been preserved is lost because a party did not take reasonable steps to preserve it and it cannot be restored or replaced. On a finding of prejudice the court may order measures no greater than necessary to cure it. Only on a finding of intent to deprive another party of the information may the court presume it was unfavorable, instruct the jury to do so, or dismiss the action or enter a default judgment.

Where do the distances in the calculation come from?

From measurements of the scene matched to the camera view: a survey with a total station or laser scanner, aerial imagery or 3D scans, plus the camera's position and a correction for lens distortion.

In the Kineticorp study the scene was measured with a total station and a Faro scanner, the camera positions were surveyed, and lens distortion was corrected in PFTrack with checkerboard calibration. Position uncertainty was defined as the distance the vehicle model could move while still matching the frame. In a road test at about 32 mph the uncertainty of the first position was plus or minus 2.7 feet, while the vehicle covered about 145 feet in 3.1 seconds, or 14 frames. The paper cites earlier studies: a maximum average error of 0.28 feet in locating evidence (Coleman and co-authors) and an average vehicle position error of 0.49 feet (Terpstra and co-authors).

Michael Jorgensen, Scott Swinford and Brian Jones of American Bio Engineers tested a camera-match overlay tool against GPS speed recorded by a RACELOGIC VBOX (SAE 2021-01-0877). They used eight camera setups: three traffic cameras, a stationary and a moving dash camera, a stationary and a moving cell phone camera, and a doorbell camera. Test speeds ran roughly 9.6 to 56.1 mph, and the authors concluded the tool was an effective method across those video sources.

Section 6 of the SWGDE court document lists what makes a vehicle look faster or slower than it was: camera angle, field of view, lens distortion, the distance of the object from the camera and lighting.

Can speed be calculated from a dashcam in a moving car?

Yes. Published methods fix a moving vehicle's position from a moving camera against measured scene data, and two SAE papers tested speed estimates from the dashcam's audio track: wheel noise over expansion joints and rumble strips, and tire tread frequency.

Engineers Manuel, Mink and Kruger of Engineering Systems Inc. considered a crash in which video from a dash camera mounted on one of the involved vehicles is available and event data recorder data is unavailable for either vehicle (SAE 2018-01-0532). They described techniques for finding the position and speed of a moving object from video taken from a moving vehicle and tested accuracy with three levels of scene data: USGS aerial photography, a total station survey, and 3D scans of the environment and the vehicles.

Henry Vega, Justin Ngo, Krystina Engleman and Jeffrey Suway of JS Forensic Consulting used the dash camera's audio track (SAE 2023-01-0632). They processed the audio in the frequency domain to remove ambient noise and calculated the test vehicle's speed as its wheels crossed expansion joints, rumble strips and grooved concrete. The authors write that the method can also give acceleration and detect turn signal activations, and they tested the audio results against video and speed data collected as a baseline.

A 2025 paper by Vega and five co-authors (SAE 2025-01-8682) tested whether a speed-dependent frequency produced by passenger car tires, which earlier research tied to the tread pattern contacting the road, can serve as a forensic method for speed. The audio was processed in iZotope RX, the speeds were compared with a VBox Sport ground-speed baseline, and the authors offer the method for use in addition to the audio methods published earlier.

How precise is a speed taken from video?

It is a range, not a single number. Its width depends on frame timing, on how well the vehicle's position can be fixed in each frame, and on how many frames separate the two positions.

The Kineticorp authors propagated the uncertainty of distance and time into speed with Taylor's method of error analysis. They took time uncertainty as two standard deviations of the frame-time error and position uncertainty as the range over which the photogrammetric match stayed acceptable. Both contributed to the speed range. In their tests position uncertainty had a minor effect, time uncertainty mattered more for the variable-rate camera, and the range narrowed as more frames separated the two positions.

The 2023 committee note to Rule 702 says each expert opinion must stay within the bounds of what can be concluded from a reliable application of the expert's basis and methodology, and that forensic experts should avoid assertions of absolute or one hundred percent certainty where the methodology is subjective.

Do US courts accept speed calculated from video?

It depends on the inputs. A federal court in Mississippi refused in 2025 to exclude a dash cam speed opinion, while an Illinois appellate court in 2024 upheld barring a surveillance video estimate built on guesses.

In Allcorn v. Beach, No. 1:23-cv-121-SA-DAS (N.D. Miss. Jan. 23, 2025), the dash cam footage from the defendant's tractor-trailer indicated that its driver was traveling 71 mph before braking where the limit was 65 mph, with 10 seconds of footage before the collision and 8 seconds after it. The plaintiff's reconstructionist, Soan Chau, tied the speed to whether the truck driver could have avoided the crash. The defense expert, Benjamin Smith, identified errors in Chau's calculations, and Chau corrected them in a supplemental report: his figure for the distance covered in 11 seconds was 1,237 feet in the first report and 1,133 feet in the supplemental one. The defendants moved to exclude him under Rule 702. Judge Sharion Aycock denied the motion and wrote that the defendants' concerns about his calculations "go to the weight of Chau's testimony, not its admissibility."

In Juliano v. Jackson, 2024 IL App (4th) 231187-U, decided July 2, 2024 as a nonprecedential Rule 23 order, two school surveillance videos did not capture the collision itself. The defense reconstructionist, Michael DiTallo, broke the nonpanning video into frames and counted one frame, or one second, between landmarks. He measured the landmark distances on an orthomosaic made of drone photographs and on 3D scans. His estimate was 21 to 25 mph about 18 to 20 feet before impact, and he later put the maximum at 28.6 mph. At his deposition he called his margin of plus or minus 2 to 3 mph an "educated guess" and did not know the exact point of impact. The trial court barred his testimony. The Appellate Court found no abuse of discretion in barring the speed opinion, because the calculations rested on speculation, quoted the rule that an expert's opinion cannot be based on mere conjecture and guess, and affirmed the judgment.

Since December 1, 2023, Rule 702 requires the proponent to show the court that it is more likely than not that the testimony rests on sufficient facts or data and is the product of reliable principles and methods reliably applied. Rule 26(a)(2)(B)(ii) requires a retained expert's report to list the facts or data the witness considered. In a video speed opinion those are the frame timing, the measured distances and the frames chosen.

How is the video itself authenticated?

Under Rule 901 the party offering it produces evidence sufficient to support a finding that the video is what the party claims it is, through a witness or through evidence about the recording system.

Rule 901(b)(1) accepts testimony that an item is what it is claimed to be, for example from the person who exported the file from the recorder. Rule 901(b)(9) accepts "Evidence describing a process or system and showing that it produces an accurate result." The advisory committee note says that example was designed for situations in which the accuracy of a result depends on a process or system.

Speed analysis often combines sources: a dashcam in one vehicle, a store camera, a traffic camera. Section 5.7 of the SWGDE court document says timing differences between combined sources should be accounted for and explained. Section 5.3 notes that one player may show the date, time and frame identification information while another player may not.

What should be sent to the reconstructionist?

Each video in its native form with metadata, details of each recording device, the scene diagram, photographs and measurements, and the vehicle data available in the case, including any event data recorder download.

Where the other side holds a vehicle or a recorder, Rule 34(a)(1)(B) lets a party request to inspect, copy, test or sample designated tangible things. The reconstructionist lists for counsel what to download and measure before the inspection.

  • The native export from each dashcam or camera system, the player or export software for a proprietary format, and a record of every copy made.
  • The make and model of each camera, its settings and where it was mounted.
  • The police diagram, scene photographs and any survey or scan of the scene.
  • Photographs of both vehicles with their make, model and tire size.
  • Any event data recorder download for the vehicles involved.

What does IAE do when speed on a video is disputed?

An IAE engineer prepares an expert report on the client's questions about crash mechanics and speed, or a review of the opposing expert's report, in up to 5 business days after the format and cost are confirmed.

The IAE accident reconstruction service covers how the vehicles moved before impact, what the event data recorder shows about speed and braking, and whether the damage on each car fits the claimed crash. The engineer matches crush on both vehicles with tire marks, gouges, the police diagram and video, and sets the point of impact, the impact angle and pre-crash speeds. IAE confirms the questions with the client before work starts.

The client sends the materials and describes the task. The secretariat confirms receipt the same day, and within one business day IAE confirms the format, the cost, the deadline and the specialist from its registry. The client approves the candidate and the CV before work starts, the draft is checked for facts, and the signed PDF is issued in up to 5 business days. A 48-hour rush is available for cases with tight deadlines.

Every specialist is a member of the public IAE registry and passed the council's attestation under the published Criteria Matrix, and the Member ID in the document can be checked by QR code. The specialist confirms in the document that he or she has no interest in the outcome of the case. Neither IAE nor the engineer guarantees the outcome, and IAE is not a law firm.

Sources

Read also

All articles

Expert examinations

Ready to discuss your case?

Send the materials: the secretariat will confirm the format, cost, deadline and the specialist within one business day.

Expertise request

Received

Cost and timeline estimate

WorkReview of an opinion
TimelineUp to 5 business days
ExpertIAE registry member