whimvention Whimsical Invention

Groundwork

When a Vehicle Notices Something, Who Decides What the Moment Means?

A new filing wants the car's cameras to turn your drive into publishable content, at the moment regulators are ordering those same cameras to forget.

Clay-diorama editorial illustration related to Information Processing System And Information Processing Method.

What changes in this design

Without this system, a reader might simply think of a vehicle detecting an occurrence and stopping there, with no stated step that turns the information into content.

Here, the job includes detecting an event inside or outside the vehicle, acquiring information connected to it, creating an editing tag, and using that tag as the basis for content.

That makes the central idea easier to spot: the record is not only about detecting an event. It also links information about that event to a label and then to content.

The dash cam that went dark in January 2020

The camera I keep thinking about stopped working. Owl Cameras sold a dash cam that plugged into the OBD-II port, recorded both inside and outside the car, and pushed the clips to a cloud service over LTE. It cost about $350. Layoffs began in September 2019, executives left in October, and the company was gone by the new year, leaving owners without support.

I keep coming back to it because that product already was the hard part. Cameras both ways, always on, connected. The sensing worked. The business didn't.

A filing that wants the drive to become content

Which brings me to US20260253615A1, Information Processing System And Information Processing Method, published this summer. What it describes is short enough to state plainly: a system detects events inside and outside a vehicle, acquires information about those events, creates an editing tag from that information, and creates content based on the tag. Not a log. Not evidence for an insurer. Content.

One thing I can't tell you is who filed it. The publication isn't indexed where I'd normally look, and three mirrors refused the request, so I won't guess at an applicant or an assignee. The mechanism is on the record; the owner isn't.

Tagging was already finished in 2017

The tagging half has been in the record for years. A commercial vehicle systems maker filed in November 2017 for a driving-oriented video recorder that reads wheel speed, acceleration, steering angle, brake pressure and yaw, adds text and face recognition on the video, and writes three partitions: the video, an index of event objects, and a light metadata tag index holding only an event type and a pointer. Its listed events include panic braking, unsignalled lane departure, a driver change detected by face, mirror checks and the radio going on and off. It was granted in 2020, and it stops at retrieval: you query the video by tag rather than scrub it.

Going the other direction is older. A 2007 filing put RFID tags along a test-drive route where GPS position triggered pre-recorded audio and video at fixed points, while cameras and microphones captured what the occupants said back. That is content written in advance and released by the drive. The new filing reverses the arrow, which is the interesting part and also the unsolved one.

Two hours of footage, and a score of 12.81

Researchers tried the reversal in 2021, at the main automotive-interface conference. Their system built road-trip summaries automatically from dashcam video plus other vehicle data, on the observation that such footage "is rarely used except for safety purposes because it takes substantial time and effort to review or edit many hours of such recorded videos", and they benchmarked it against summaries their participants cut by hand. Five years on, the paper has four citations.

There's a reason nobody followed. Highlight detection is measured on short video: one standard set of 50 clips averages 4.2 minutes per video; a much larger one averages 3.4. Against 320 videos averaging two hours each, the best method managed 12.81 mAP. A car makes two-hour videos every day of its life.

Where the risk actually sits

The law has spent that same period pushing this hardware the other way. Commission Delegated Regulation (EU) 2023/2590 requires distraction warning on new vehicle types from 7 July 2024 and on all new cars and vans from 7 July 2026, about a month before this filing published. It watches where the driver looks and warns within six seconds above 20 km/h - and it must do that without biometric information, as a closed loop keeping only what the system needs to work. The American equivalent, finalised in December 2024, widens crash capture from five seconds at 2 Hz to twenty seconds at 10 Hz. Twenty seconds.

The cost of ignoring that shape is documented. Samsara settled a class action for $3.95 million over a dash cam feature alleged to have captured drivers' facial geometry in Illinois without consent. Reporting in April 2023 described Tesla employees privately circulating footage from customers' cars between 2019 and 2022, including scenes inside homes and garages, while company material told owners their data belonged to them.

So: the editing tag isn't the invention, and it isn't the danger either. Tagging is done, the sensing shipped and failed commercially six years ago, and the summarising is unsolved at the length a commute produces. What is new is purpose. A crash recorder and a distraction warning are built to keep almost nothing. A content pipeline is built to keep whatever turns out to be good. Point the same lens at the same passenger with the second purpose and every question the first purpose had settled - what's kept, who sees it, who is liable when it leaks - opens again.

Sources

Reporting for this piece drew on the published filing and the references below.

  1. US20260253615A1 - Information Processing System And Information Processing Method (Justia)
  2. Automatic Generation of Road Trip Summary Video for Reminiscence and Entertainment using Dashcam Video (AutomotiveUI 2021)
  3. SVHighlights: Towards Extremely Long Sport Video Highlight Detection
  4. EU Regulation on Advanced Driver Distraction Warning Systems Published
  5. Understanding Advanced Driver Distraction Warning (ADDW) systems
  6. USA - Event Data Recorders (49 CFR Part 563)
  7. $3.95M Samsara Dash Cam Settlement Ends Class Action Over Alleged Biometric Privacy Violations
  8. Tesla workers shared sensitive images recorded by customer cars
  9. Owl Cameras is dead, and users who bought its expensive dash cam are in limbo
  10. Mozilla flunks 25 major car brands for data privacy fails

The record, briefly. US20260253615A1, “Information Processing System And Information Processing Method” An information processing system detects events inside and outside a vehicle, acquires event-related information, creates an editing tag from that information, and creates content based on the editing tag.

More angles on this

Read around this one:Public patent recordUSPTO published figure