Capture and ingestion
Upload or live capture with frame rate, resolution and orientation normalised, and metadata recorded so a later reanalysis is reproducible.
Solution
Computer Vision and Video Analytics
Applications that extract measurements from images and video: tracking movement, reading conditions, and turning frames into numbers a person can act on.
Overview
What the system does
Computer vision is useful when the information you need already exists as an image and the alternative is a person watching and writing things down.
The work divides into detection, tracking and measurement. Detection finds the thing in a frame, tracking follows it across frames, and measurement turns the trajectory into a number with units.
The measurement step is where most projects succeed or fail, because a pixel path is only meaningful once it is calibrated against something real.
Who needs this
Teams extracting useful observations from images or recorded video, with a defined tolerance for uncertainty.
What we build
Separate detection, tracking and measurement. A pixel trajectory needs calibration before it becomes a physical measurement. Evaluate lighting, camera angle, occlusion and device quality with representative footage before choosing an approach.
Modules
Core modules
Detection
Object, marker or pose detection per frame, chosen against the accuracy the measurement needs rather than the most elaborate model available.
Tracking
Association of detections across frames into a continuous path, with smoothing and gap handling when a frame misses.
Calibration and measurement
Pixel to real world scaling from a known reference, so outputs carry units. Without calibration a path is a picture, not a measurement.
Replay and visualisation
Overlays synchronised to the video, scrubbing, frame stepping and side by side comparison between attempts or time periods.
Derived metrics
Velocity, displacement, angle, timing, count and dwell, computed from the tracked path and defined explicitly.
Processing pipeline
Queued jobs with progress reporting, retries and a cost ceiling, because video processing is slow and expensive enough to need both.
Privacy handling
Retention rules, access control, and blurring or discarding of identifiable material where the use case does not require it.
Architecture
How it fits together
A typical implementation flow, with explicit responsibilities at each boundary.
Integrations
- Object storage for source video and rendered output
- Processing services with hardware acceleration
- Client applications for capture and playback
- Operational systems that consume the measurements
- Notification services for completed analysis
- 01
Ingest
Validate footage, orientation, duration and processing consent.
- 02
Analyze
Run bounded detection and tracking jobs with versioned model settings.
- 03
Review
Show source footage alongside results, uncertainty and controls for human interpretation.
Security and controls
Design for the exceptions, not just the happy path.
Restrict identifiable footage, define retention and keep upload access private. Rate-limit expensive jobs and cap processing cost. Do not claim biometric or safety accuracy that evaluation has not established.
Technology options
We select the final stack after reviewing your existing systems and operating requirements.
Security and IP practicesFAQ
Questions people actually ask
How accurate is the measurement?
It depends on capture conditions, frame rate and calibration, and it is measured against known references during the build rather than claimed in advance. We report the accuracy we achieve on your footage.
Can it run on a phone?
Detection and tracking can run on device for short clips on current hardware, which removes upload time and keeps footage local. Longer or higher resolution work is better queued to a server.
Why does calibration matter so much?
Without a known reference in the frame, the system can tell you a path in pixels but not in centimetres. Calibration is what turns a picture into a measurement.
How is video stored and for how long?
In object storage you own, with a retention policy set per use case and signed short lived links for playback. Footage is not kept longer than the purpose requires.
Does this involve facial recognition?
No. We do not build facial recognition or identification systems. The work here is movement, object and condition measurement.
How long does analysis take?
It scales with clip length, resolution and the model used. Jobs are queued with progress reporting and a cost ceiling so a long upload never runs away.
Let’s talk about the system you need.
Tell us where the current process breaks down. We can map the workflow, examine the integration requirements and define a practical scope for the build.
- A mutual NDA is signed before project details are shared.
- Scope is fixed in writing before any build starts.
- You own the source code, the IP and the documentation at delivery.
- Support plans are available and tailored on a discovery or technical call.