Turn your video system into a profit center.
Your video feeds into a proprietary 3D model, where each visitor becomes a single anonymous path - precisely geolocated across fields of view, occlusions, and camera handoffs.
Use Your Existing Video
Get more value from the video you already have - without installing new cameras, on-site servers, or additional software.
Connect Your Footage
Upload video directly, connect your VMS, or securely archive footage through our partner, Wasabi.
Process It with Onsights AI
Our proprietary AI detects, tracks, re-identifies across cameras, and precisely maps foot steps.
Unlock Onsights Telemetry
Use movement telemetry to create actionable insights across everywhere you have cameras.
Analytics UI
Quickly review KPIs, insights and trends for total situational awareness.
Data APIs
Grab the raw data and run your own analysis to drive decision making. Included with an Enterprise license.
Reporting
Experiment with A/B testing, causal modeling, event correlations and more.
Encrypted in transit & at rest · no PII, no facial recognition · your video is deleted after processing.
Thirty minutes of your time. Live within two days.
Your part · about 30 minutes
- 1 Enter facility details - with AI assist
- 2 Connect your data
- 3 Select your cameras
- 4 Submit & done
Our part · under 2 days
We do the rest.
We run our proprietary 3D model, tune our system to your facility, validate and email your confirmation and login within two days.
No specialized hardware, no survey, no new cameras, no servers to maintain - and no hidden costs.
Quality
Onsights provides measured results under both recommended and challenging conditions. Our telemetry gracefully handles when facility conditions push past recommended conditions. Results vary by scene geometry, camera placement, and crowd density.
The numbers below reflect the system as deployed, not a controlled lab environment.
Region Visitor Counts
A region visit is recorded each time a visitor enters a user-defined area: a store, a department, an aisle, an entrance, a display. We report unique visitor counts, counting each person once per region regardless of how many times they enter. Repeat visits can be computed separately, so you know both how many distinct visitors reached an area and how many came back.
0.99
Recommended conditions
0.94
Challenging conditions
Agreement between system-counts using Onsights telemetry and human annotator-counted visits per region.
- Region
- Any user-defined area of interest - a store, a department, an aisle, an entrance, a display - drawn on the facility floor plan.
- Recommended conditions source
- Sample taken from 11 cameras at a major luxury mall during a typical day that had approximately 25,000 visitors.
- Methodology
- System-counts created from Onsights telemetry are validated against human-annotated ground truth. The annotators independently reviewed, with replication, thousands of camera-hours and labeled the relevant events at the frame-level; system outputs were compared against these labels. We report results as measured - no cherry-picking of scenes or time windows.
- Challenging conditions
- Sample taken from a super regional outlet mall during peak hours of a Black Friday week day, using over 140 cameras on a day with over 80,000 visitors.
Operating conditions
Our quality depends on the quality and configuration of your existing camera system. The factors below have the largest influence on performance - and determine where results fall on the spectrum between our recommended-condition and challenging-condition benchmarks.
Camera coverage
Gaps between cameras create gaps in visitor journeys. Near-complete coverage of the facility is important for accurate path and dwell analytics.
Video resolution
Full HD is the minimum. Higher-resolution cameras improve detection reliability, particularly in dense or distant scenes.
Frame rate
Adequate frame rate is required for motion continuity. Very low frame rates break tracking between frames and reduce cross-camera accuracy.
Camera angle
Very steep (near-overhead) and very shallow (near-horizontal) viewing angles reduce positional accuracy. Cameras mounted at a moderate overhead angle perform best.
Visitor volume
Extreme crowd density increases occlusion and identity errors. Our challenging-condition results show what degradation looks like at peak conditions.
Camera stability
Our system is calibrated to each camera's fixed position. Bumped or shifted cameras degrade results until recalibration is performed.
Clock synchronization
Cameras must maintain consistent time alignment with one another. Clock drift between cameras creates cross-camera tracking artifacts.
Facility changes
Seasonal obstructions, new fixtures, or floor layout changes affect accuracy. Let us know when significant changes occur so we can update the spatial model.
Unsupported inputs: Fisheye and heavily distorted lenses are outside our operating envelope. PTZ cameras that pan, tilt, or zoom during operation, and cameras with proprietary on-device video processing or stitching transforms, are also not supported.