§ 00 — Computer vision · Human movement

Teaching a single camera to understand how the body moves.

TZ Martin has spent years building computer-vision systems that quantify human movement from ordinary video — markerless motion capture for clinics and athletes, where the only sensor is a camera.

01In Brief

TZ Martin is a computer-vision engineer and founder who builds systems that quantify human movement from a single camera. Using pose estimation and deep learning, his work turns ordinary video into clinical-grade motion data — no markers, no motion-capture suits, no specialized hardware.

The technical center of that work is markerless motion capture: detecting body keypoints frame by frame, reconstructing joint angles and kinematics, and analyzing the physics of how a person actually moves. It is computer vision applied to biomechanics — the same discipline behind both FDA-cleared clinical motion analysis and real-time athletic performance tools.

02Track Record
01

DARI Motion — Scientific Analytics

As CTO, rebuilt DARI Motion into a real-time video-analytics platform and co-led its 510(k) FDA clearance for markerless human-motion analysis — full-body assessment from camera input, with validated results returned to the cloud.

02

Virtruvia Systems

Founded a single-camera video-analytics platform applying pose estimation and deep learning to record, track, and report on human movement for athletics and health — localized AI running on a mobile device.

03

Pose estimation & kinematics

Work centered on human pose estimation and the causal analysis of kinematics — quantifying biomechanics and the physics of movement from a single point of view, an area TZ has focused on since 2017.

04

Toolchain

Built on OpenCV, TensorFlow, PyTorch, and CoreML, with real-time pipelines on Google Cloud — the same stack used across his AI and data-engineering work.

03Detail

What markerless motion capture does

Traditional motion capture needs reflective markers, body suits, and a dedicated camera rig. Markerless motion capture replaces all of that with computer vision: a model locates the body's joints in each video frame, then reconstructs how those joints move through space over time. The output is the same kind of quantified movement data — joint angles, velocities, asymmetries — from nothing but video.

Why a single camera is hard, and worth it

Estimating three-dimensional movement from a flat, single-camera image is an under-constrained problem: depth, occlusion, and perspective all have to be inferred. Solving it well removes the cost, calibration, and lab requirement of multi-camera systems — which is what lets motion analysis move from a research lab into a clinic, a gym, or a phone.

From movement data to decisions

Quantified movement is only useful if it changes a decision. In clinical settings that means objective, validated measurements clinicians can act on; in athletics it means tracking performance and injury risk over time. The throughline across TZ Martin's computer-vision work is turning raw video into evidence — the same fragmentation-into-agency thesis that runs through his ventures.

04Questions

What is markerless motion capture?

Markerless motion capture uses computer vision to measure human movement from ordinary video, without reflective markers, body suits, or a multi-camera rig. A model detects the body's joints in each frame and reconstructs the motion, producing quantified data such as joint angles and movement asymmetries.

How does computer vision measure human movement from a single camera?

A pose-estimation model locates body keypoints in every video frame, and deep-learning methods infer three-dimensional motion from that two-dimensional input. The system then reconstructs joint angles and kinematics over time, turning a single-camera video into structured movement data — the approach TZ Martin built at Virtruvia Systems and DARI Motion.

What has TZ Martin built in computer vision?

TZ Martin founded Virtruvia Systems, a single-camera video-analytics platform, and as CTO rebuilt DARI Motion (Scientific Analytics) into a real-time motion-analysis platform, co-leading its 510(k) FDA clearance. Both apply pose estimation and deep learning to quantify human movement for healthcare and athletics.

What is pose estimation used for in healthcare and sports?

Pose estimation quantifies how a person moves, enabling objective biomechanics assessment without lab equipment. In healthcare it supports clinical motion analysis and rehabilitation tracking; in sports it measures performance, technique, and injury risk. It is the core technique behind markerless motion capture.

What tools does TZ Martin use for computer vision?

TZ Martin's computer-vision work is built on OpenCV for image processing, TensorFlow and PyTorch for deep learning, and CoreML for on-device inference, with real-time data pipelines running on Google Cloud.

06Continue

Quantify what was once invisible.