Video near miss analysis: what traffic conflict studies show and what they cannot predict
How close call studies work, how to judge them and why they are not forecasts
Safety Behind the Wheel Foundation built this guide from the Federal Highway Administration (FHWA), the National Highway Traffic Safety Administration (NHTSA) and published research on traffic conflicts, video analysis and naturalistic driving. Every source is listed at the end.
What video conflict analysis is
A traffic conflict is a close approach between two road users. If neither one brakes or swerves, they will collide. People also say near miss, near crash or close call. The terms overlap in everyday use.
Video conflict analysis turns that idea into a routine. Cameras record a place for many hours. Computer vision, which is software that reads video, finds each vehicle, rider and walker and follows their paths. It measures how close each pair came. Trained people review the flagged clips and keep the real ones.
Researchers call the results surrogate safety measures. A surrogate is a substitute. Close calls stand in for collisions that have not happened [1].
Start with a question, not a camera. For example: how often do drivers turning left come very near people in the crosswalk at this corner? A narrow question shows where to point the camera and what to count.
This is not an enforcement camera. A speed or red light camera is built to issue citations. Analysis footage is meant to study patterns.
Why look at near misses at all
Crashes are a slow teacher. Reviewers note that they are rare, so a site needs a long watch before numbers mean much. That creates a hard choice: wait for enough collisions, some causing injury or death, before acting. Police crash data also has gaps, such as underreporting and little detail about behavior [1].
FHWA makes a related point: serious collisions rarely bunch up at one spot over time. They seem to land at random places, yet the risk factors behind them are very similar [2]. Looking only where crashes happened before can miss the next dangerous place.
The Safe System approach is FHWA’s framework for reaching zero deaths. It rests on the ideas that people make mistakes and that safety work should be proactive [9]. Proactive means acting before harm. Close call video is one way to see mistakes while they are still harmless.
The stakes are real. NHTSA’s website reported 39,254 lives lost on U.S. roads in 2024 [10]. Finding a hazard before it hurts a family is the goal.
The measures, in plain words
Most studies use a few common measures. Each one asks, in a different way, “How close was that?”
Comparison compiled by Safety Behind the Wheel Foundation
Common traffic conflict measures
| Measure | Plain meaning | What it tells you |
|---|---|---|
| Time to collision (TTC) | Seconds left before two road users would collide if both held speed and path | Less time means a closer call [11] |
| Post encroachment time (PET) | Seconds between one road user leaving a spot and the next one arriving | A smaller gap means a tighter near miss [5, 11] |
| Approach speed, angle and distance | How fast each party moved, the angle between their paths and the closest they came | Helps describe how the event unfolded [11] |
| Evasive action | Hard braking or swerving to avoid contact | Shows someone had to react, and helps tell collisions from near misses [7] |
The field has no single, widely agreed rulebook. One review of 549 studies set out to build a unifying framework because none existed [1]. In Norway, three methods were applied to video from three intersections. All were compatible with crash records. The two manual methods agreed fairly well, but some differences had no obvious explanation [3]. So always ask which measure and which cutoffs a report used.
Where it helps: intersections, bus stops, yards and crossings
Intersections are the classic use. A Pennsylvania Department of Transportation study used a video based event monitoring system. It looked at interactions between motor vehicles and people walking or biking at 15 intersections with traffic signals [4].
Midblock crossings, which lie between intersections, can be compared too. A study of Florida crossings used portable cameras and computer vision to count conflicts at 14 sites with a newer midblock signal. The authors reported 26% to 33% fewer serious conflicts than at pedestrian hybrid beacons, which are signals that stay dark until someone asks to cross [12].
School bus stops are a natural fit. NHTSA’s stop and route toolkit helps districts review stops, answer parent requests and plan new ones. It takes a Safe System view that expects human mistakes [13]. A district could add short video from a stop at pickup time. It might show where children wait and where close calls repeat. That adds evidence without replacing the toolkit.
Yards and depots, such as bus lots, truck yards and loading docks, mix backing vehicles with people on foot. A camera at a yard entrance can show where close calls repeat. The studies in our sources covered public roads, so ask for proof that a tool works at your site. Our guides on backing safety and work near miss investigation cover the human side.
Railroad and road crossings work the same way. A camera can log how often drivers stop on the tracks or weave around gates. The agencies and railroad that own a crossing decide on changes. Our railroad crossing guide covers the warning devices.
Road agencies, school districts, fleet managers, nonprofits and university teams run most studies. Parents and everyday drivers can still ask for one.
How a study runs, step by step
Follow these steps in order.
- Write one narrow question and name the decision the answer will inform.
- Pick sites and times. Include busy and quiet hours, daylight and dark. Studies in our sources used tens of hours of video, such as over 51 hours at one busy intersection [11] and 64 hours across three traffic environments [5].
- Get permission. The road or property owner should approve where cameras go. Portable units exist for flexible data collection [11].
- Record, then run detection. Software finds each road user, follows its path and flags close events.
- Have trained reviewers confirm every flagged clip and discard false alarms.
- Turn counts into rates using traffic volume, such as conflicts per 1,000 left turns. Without volume, a busy site always looks worse.
- Compare sites, or compare before and after a change, keeping camera angle, hours and settings the same.
- Send the findings to the road owner with a clear request.
Review detection quality before you trust a number
Detection quality is the weak link. Software must see a person, name what it is and follow it frame by frame. Small position errors become errors in the close call measures. One 2026 study shows how much. Most video pipelines mark each person or vehicle with a box and treat the center of the box as its location. A more realistic point changed post encroachment time by 0.69 seconds on average. Results differed by more than one second in 16.7% of conflict events, and up to 3.6% of events landed in a different risk category [5].
Poor tracking also creates false alarms. In the Norway comparison, the automated method reported many more interactions than the manual methods, including milder ones. The authors warn that inaccurate paths from the video may have produced a significant share of false alarms [3].
Human review matters too. Researchers building on the Pennsylvania data tested whether automatic measures could predict conflicts that people had confirmed, with no person in the loop. Some measures were more useful than others, and the mix differed for bicycle and pedestrian conflicts [4]. That study also tracked lighting and weather. In plain terms, automatic flags are a first pass, so ask whether accuracy was tested in the conditions you face.
Near misses are clues, not forecasts
Naturalistic driving studies give one test of that idea. They put cameras and sensors in participants’ vehicles. In the 100 Car Study, 90 drivers were followed for 12 to 13 months. For run off road events alone, researchers counted 14 crashes and 182 safety related events, including 30 near crashes, and found positive links among them [6]. That supports using close calls as a signal.
The same paper lists long standing criticisms of this approach: poor data quality, missing measures of exposure and the difficulty of judging fault [6]. Exposure means how much traffic or walking passed by.
A later large scale naturalistic study found that crashes and near crashes differ in important ways. What the driver was doing before the event, and the evasive move the driver made, helped tell them apart. The authors still say surrogate events give useful information for preventing collisions [7].
Severity matters. One study looked at left turning drivers and oncoming traffic at signal controlled intersections, a well known source of serious injury collisions. Models built on the most extreme conflicts fit crash counts better than models built on all conflicts. The gain was larger for fatal and serious injury collisions [14]. So a long list of mild close calls can mislead.
A large review calls for more work to build solid links between conflicts and collisions [1]. Until then, treat each result with care. Common mistakes include:
- Using close calls to forecast crash counts, instead of deciding where to look.
- Ranking sites by raw counts, without traffic volume or severity.
- Skipping several years of serious injury and fatal crash data. Our guide on using crash data for community priorities shows how to use it.
- Comparing unlike hours, seasons or traffic.
- Trusting small samples, where a few events can swing a ranking.
- Recording with no plan to act on the results.
Getting the service: who offers it and how to choose
Buyers are usually road agencies, school districts, fleets and nonprofits. Parents and other consumers rarely buy this. They request it or help fund it. The usual channels are:
- City, county and state transportation offices. Many hold video or run short studies. Start with the traffic engineering office.
- University transportation research centers, which often partner with local agencies.
- Engineering firms and analysis vendors that sell video analytics, including portable camera kits [11].
- Public data tools, such as crash dashboards and open data portals. They show collisions, and rarely close calls.
- Fleet safety service providers that supply in vehicle cameras. Our guide on fleet telematics and driver cameras covers coaching.
- Training providers and driving schools, which may use deidentified clips in lessons. Ask how they handled consent.
Match the tool to the question. A one time study needs a written method. Ongoing monitoring needs steady upkeep and someone who reads the results. Ask for sample reports, similar clients and exportable data. Be wary of any claim that a product predicts collisions.
Privacy, fairness and upkeep
Video shows faces, plates and children. Researchers describe data protection as a challenge when collecting behavior video [11]. Good habits:
- Collect only what the question needs, and blur faces and plates when you can.
- Limit who sees raw video, and keep a list.
- Set a delete date and follow it.
- Tell the community what is recorded and why.
- Keep safety review separate from discipline.
- Check local privacy and public records rules with the agency’s counsel. Rules differ by place and change over time.
Put cameras where the risk is, not only where mounting is easy. Camera systems drift. Lenses get dirty, mounts shift and software updates change results. Check a sample of clips monthly against what a person sees. Log every camera move and software change. Keep settings fixed during a before and after comparison.
Turning findings into action
Only authorized agencies may select, design and install traffic control devices on public roads. Residents should never paint crosswalks or post signs. Send findings to the road owner instead.
One good next step is a road safety audit. FHWA describes it as a review by a multidisciplinary team that is independent of the project. The team considers all road users and writes a formal report, and the road owner must respond [8]. FHWA’s page cites a 10% to 60% reduction in total collisions for this countermeasure [8]. Safety Behind the Wheel Foundation encourages communities to ask their road owner about one.
FHWA’s list of proven safety countermeasures includes crosswalk visibility enhancements, leading pedestrian intervals and pedestrian hybrid beacons. It also includes dedicated turn lanes, roundabouts and yellow change intervals (how long the yellow light lasts) [15]. A leading pedestrian interval gives people on foot a head start before turning drivers get a green. Engineers decide whether any fits your site.
For bus stops, ask the district transportation office for a stop review using the NHTSA toolkit [13]. For yards, employers can pair video with driver reports and adjust traffic flow and walkways.
Frequently asked questions
How many hours of video does a study need?
We found no single national rule in the sources we reviewed. Examples include over 51 hours at one busy intersection [11] and 64 hours across three traffic environments [5]. The right amount depends on how often interactions happen and which times of day matter, so ask the analyst to explain the choice.
Can software find near misses without a person checking?
Not reliably. Software can flag candidates, but in one test automatic flags were compared against conflicts that people confirmed, and some measures worked better than others [4]. Inaccurate paths also caused false alarms [3]. Treat software as a first pass.
Is this the same as a speed or red light camera?
No. Enforcement cameras exist to cite drivers. Analysis cameras collect footage to study how road users interact. Ask the program owner whether footage can be used in other ways. Our guide on speed and red light camera laws by state explains enforcement cameras.
Can I use my own dash camera clip to ask for a safety review?
A clip can support a request but shows one event. Note the date, time and place, send it to the road owner, and do not post it online. Our guide on saving dash camera video after a crash covers protecting the original file.
What should a school board or small employer ask first?
Start with the question: what decision will the results change? Then ask what data you already have, who will review flagged clips, how accuracy was tested, who owns the footage, and who will act on the results.
The bottom line
Recommendation from Safety Behind the Wheel Foundation
Safety Behind the Wheel Foundation encourages communities to use close call video to find risky places early, while checking how each result was made. Video conflict analysis lets a community, school district or employer see where road users nearly collide, before someone is hurt. It works best as a clue that points to places worth a closer look. It works poorly as a collision forecast.
Ask how close calls were detected and who checked them. Pair the results with serious injury and fatal crash data. Protect privacy, and send findings to the people who own the road.
Related articles
- How to use local crash data to choose community safety priorities
- Safe System approach: how road design protects you from human mistakes
- Work crash and near miss investigation: a step by step employer guide
- Fleet telematics and driver cameras: how to coach drivers, not punish them
- School bus and school zone safety: when to stop and how to protect kids
Sources and further reading
- 1Arun A, Haque MM, Bhaskar A, Washington S, Sayed T. A systematic mapping review of surrogate safety assessment using traffic conflict techniques. Accident Analysis and Prevention, 2021, vol. 153 (via PubMed; doi:10.1016/j.aap.2021.106016). Data year: review of 549 studies (study years not stated in abstract). pubmed.ncbi.nlm.nih.gov/33582529
- 2Federal Highway Administration. Why the Systemic Approach? Last updated February 18, 2026. highways.dot.gov/…/why-systemic-approach
- 3Laureshyn A, Goede M, Saunier N, Fyhri A. Cross-comparison of three surrogate safety methods to diagnose cyclist safety problems at intersections in Norway. Accident Analysis and Prevention, 2016, vol. 105 (via PubMed; doi:10.1016/j.aap.2016.04.035). Data year: video from three Norwegian intersections (dates not stated in abstract). pubmed.ncbi.nlm.nih.gov/27290729
- 4Sengupta A, Ilgin Guler S, Gayah VV, Warchol S. Evaluating the reliability of automatically generated pedestrian and bicycle crash surrogates. Accident Analysis and Prevention, 2024, vol. 203 (via PubMed; doi:10.1016/j.aap.2024.107614). Data year: video from 15 Pennsylvania signalized intersections (dates not stated in abstract). pubmed.ncbi.nlm.nih.gov/38781631
- 5Janoška Z, BĂlová M, BĂl M. A novel method for determining the trajectory of vulnerable road users from video recordings and its practical implications. Accident Analysis and Prevention, 2026, vol. 233 (via PubMed; doi:10.1016/j.aap.2026.108578). Data year: 64 hours of video from three traffic environments (dates not stated in abstract). pubmed.ncbi.nlm.nih.gov/42105696
- 6Wu KF, Aguero-Valverde J, Jovanis PP. Using naturalistic driving data to explore the association between traffic safety-related events and crash risk at driver level. Accident Analysis and Prevention, 2014, vol. 72 (via PubMed; doi:10.1016/j.aap.2014.07.005). Data year: 12 to 13 months of 100-Car Study driving by 90 drivers (exact years not stated in abstract). pubmed.ncbi.nlm.nih.gov/25086439
- 7Perez MA, Sudweeks JD, Sears E, Valente J, Guo F. Differences in frequency of occurrence, event characteristics, and pre-impact vehicle kinematics between crashes, near-crashes, and single vehicle conflicts in a large-scale naturalistic driving study. Traffic Injury Prevention, 2022, vol. 24 (via PubMed; doi:10.1080/15389588.2022.2155785). Data year: Second Strategic Highway Research Program Naturalistic Driving Study events (years not stated in abstract). pubmed.ncbi.nlm.nih.gov/36548218
- 8Federal Highway Administration. Road Safety Audit (Proven Safety Countermeasure page). Accessed October 2026 (no update date shown). Data year: not stated on the page (it cites reports FHWA-SA-12-037 and FHWA-SA-06-06 for the crash range). highways.dot.gov/…/road-safety-audit
- 9Federal Highway Administration. Zero Deaths and Safe System. Last updated August 6, 2025. highways.dot.gov/…/zero-deaths-and-safe-system
- 10National Highway Traffic Safety Administration. Home page statistic, 39,254 lives lost on U.S. roads in 2024. Accessed October 2026. Data year: 2024. nhtsa.gov
- 11Hula A, Ziakopoulos A, Losada A, Schaub A, Saleh P, Yannis G. Trajectory-based indicators to determine the local character of intersection conflicts: A micro-spatial analysis. Accident Analysis and Prevention, 2025, vol. 220 (via PubMed; doi:10.1016/j.aap.2025.108155). Data year: over 51 hours of video at one Vienna intersection (dates not stated in abstract). pubmed.ncbi.nlm.nih.gov/40616943
- 12Ahsan MJ, Abdel-Aty M, Abdelrahman AS. Can mid-block pedestrian signals (MPS) provide greater safety benefits than other mid-block pedestrian crossings? Accident Analysis and Prevention, 2025, vol. 218 (via PubMed; doi:10.1016/j.aap.2025.108105). Data year: portable video at 14 Florida signal sites plus 5 reference sites (dates not stated in abstract). pubmed.ncbi.nlm.nih.gov/40373590
- 13National Highway Traffic Safety Administration. Planning Safer School Bus Stops and Routes. Accessed October 2026 (toolkit created in 2022; no update date shown). nhtsa.gov/…/planning-safer-school-bus-stops-and-ro…
- 14Hasanpour M, Persaud B, Mansell R, Milligan C. Investigation of a surrogate measure-based safety index for predicting injury crashes at signalized intersections. Traffic Injury Prevention, 2024, vol. 26 (via PubMed; doi:10.1080/15389588.2024.2397652). Data year: video-derived conflicts and crashes at signalized intersections (years not stated in abstract). pubmed.ncbi.nlm.nih.gov/39325686
- 15Federal Highway Administration. Proven Safety Countermeasures. Accessed October 2026 (no update date shown). highways.dot.gov/…/proven-safety-countermeasures
This article reflects the safety priorities of traffic safety data and proactive road safety work. It is general education, not legal or professional advice. Methods and rules change, so verify details with official sources. For your own situation, contact your local road agency, school district transportation office or employer safety manager, and ask a privacy or legal professional about video rules. Reviewed October 2026.
Cite this article: Safety Behind the Wheel Foundation. “Video near miss analysis: what traffic conflict studies show and what they cannot predict.” drivewithcare.org, https://drivewithcare.org/video-near-miss-conflict-analysis/. Reviewed October 2026.
Special thanks to our contributors
We appreciate the professionals, experts, and community members who share their knowledge and perspectives to help make this resource better. Your insights are always welcome and appreciated.
Have professional expertise, firsthand experience, or a perspective that could make this resource better?
We welcome your input. Your knowledge and experience can help us make this resource more useful, accurate, and meaningful to the people who rely on it.
Thank you. Your contribution is on its way.
We’re grateful you took the time to share your knowledge with us. Our team reviews every contribution with care. If it helps improve this resource, we’ll update the page and recognize you the way you chose.
Keep learning
Safe Driving Resource Center
Guides, self-checks and articles for drivers, parents of new drivers, and families recovering after a crash.
