beginner
safety
computer-vision
helmet

Levy Vision Overview

What the Levy Vision computer-vision safety suite does, what it covers today, and where it's headed

Levy Fleets TeamMay 18, 20266 min read

Levy Vision Overview

Levy Vision is the computer-vision safety suite built into Levy Fleets. It uses AI to check that riders are wearing helmets, validates parking-pose at the end of every ride, ingests sidewalk-detection events from on-vehicle vendor hardware, and lets you enforce policy automatically — from a soft in-app warning all the way to throttle cut.

This article gives you the lay of the land. The rest of the Safety & Vision category goes deep on each capability.

Who this is for

Levy Vision is designed for three different audiences, all working from the same dashboard:

  • Operators who want lower crash liability, fewer sidewalk complaints, and a defensible insurance story
  • Field techs who need to clear false positives and restore throttle on a misclassified ride
  • City compliance officers who require an MDS-aligned export of safety events for permit reporting

You access everything from Dashboard > Safety in the sidebar.

What's in the suite

Levy Vision ships in three layers. You can turn each one on independently per subaccount.

Helmet verification at unlock

Before a ride starts, the rider takes a selfie. Our classifier checks for a visible helmet and either passes the unlock, asks for a retake, or blocks the ride if you've enabled strict mode. Helmet-lock hardware (such as an AXA helmet lock) can bypass the selfie when the helmet has been freshly unlocked.

Read more in Helmet verification at unlock.

Parking-pose validation

At the end of every ride, the rider's parking photo is classified for pose (upright vs tipped), zone compliance, and whether it blocks the sidewalk. A high-confidence fail can trigger a reparking fee; medium-confidence prompts a retake; clean passes do nothing. Riders can appeal with a second photo, and you resolve the appeal from the operator dashboard.

Read more in Parking-pose validation.

Sidewalk detection (with throttle cut)

For fleets running on-vehicle hardware from Drover AI or Luna Systems, sidewalk-riding events stream into Levy Fleets in real time. A policy ladder you control turns those events into action: warn the rider, reduce top speed, or cut throttle entirely.

Read more in Sidewalk detection and Throttle-cut policy.

Dashboard surface

Once enabled, you get a new Safety group in the sidebar with these pages:

PageWhat it's for
OverviewTop-line metrics — helmet pass rate, parking pass rate, sidewalk events per 100 rides
EventsUnified feed of every CV event across your fleet (filterable, exportable)
Helmet complianceHelmet selfie pass / fail / ambiguous trends
Parking reviewsAppeal queue — operator action required
Sidewalk hotspotsMap of where sidewalk events cluster (city-permit gold)
Safety settingsPer-subaccount toggles — strict mode, fee schedule, thresholds
MDS exportCity-compliant safety event export by date range and polygon

Off by default

Every Levy Vision feature ships off by default. Nothing in this category affects your rides until you flip the relevant toggle for a subaccount in Dashboard > Safety > Safety settings.

How accuracy is governed

We hold ourselves to published targets that match the spec:

  • Helmet verification: 92%+ true-positive rate, under 5% false-positive rate
  • Sidewalk detection: 85%+ precision (false positives are operationally expensive, so we tune conservative)
  • Parking-pose: 90%+ upright-detection accuracy across scooter, bike, e-bike, and moped

Every event is logged with a confidence band (high, medium, low) so you can decide where the bar sits for your fleet.

What you'll need to get value

  • A subaccount with at least one fleet of vehicles (any OEM works for helmet + parking)
  • For sidewalk detection: a vendor agreement with Drover AI or Luna Systems (see Vendor setup)
  • For city compliance: a defined jurisdiction polygon and a city contact who consumes MDS feeds

Where to go next