What is it about?
Our review provides a detailed examination of the recent advances in markerless motion capture systems and the prior research that made it possible. It focuses on methods capable of tracking multiple people in real- time using multiple cameras, without the need for physical markers. The review systematically analyzes the performance of state-of-the-art approaches in terms of accuracy, latency, and computational efficiency. It explores key architectural designs such as top-down, bottom-up, and voxel- based pipelines, and evaluates how these influence the aforementioned metrics. A comprehensive analysis of the performance of keys methods on modern GPU helps understand the scaling law with the number of person tracked in a scene and the number of cameras that are needed to accurately track motion. By mapping the field’s evolution from early geometric methods to modern deep learning–based systems, the paper identifies both major breakthroughs and the remaining challenges towards fully markerless, real- time 3D motion tracking in multi-person scenarios.
Featured Image
Photo by Aditya Enggar Perdana on Unsplash
Why is it important?
Real-time markerless motion capture creates a path toward more natural, accessible, and scalable motion analysis. By eliminating the need for physical markers, it enables seamless motion tracking in real-world settings, examples of which range from sports or physical rehabilitation to immersive virtual and augmented reality experiences. Real-time performance enables realistic interactions and feedback. Markerless motion capture stands at the intersection of computer vision, biomechanics, and machine learning, driving innovation in deep learning– based pose estimation, multi-view fusion and real-time 3D reconstruction. Advancements in this field not only improve the accuracy and efficiency of motion analysis technologies but also strengthen our understanding of how to model and interpret complex human motion at scale.
Perspectives
This article was a deep-dive into the computer vision methods applied to motion-analysis. Understanding the scaling low and the required computing of past and current methods helped to identify the promising methods. Measuring and understanding the world in real-time is a requirement for any system that interacts with human.
Dr Pierre Nagorny
Artanim Foundation
Read the Original
This page is a summary of: A Comprehensive Review of Real-Time Multi-View Multi-Person Markerless Motion Capture, ACM Computing Surveys, September 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3757733.
You can read the full text:
Resources
Markerless Mocap
The Markerless Mocap project aims to leverage modern Machine Learning (ML) approaches to create a markerless solution for the capture of users in LBVR scenarios. The low-latency requirements imposed by this scenario are the primary challenge for the project, with an ideal photon-to-skeleton latency being in the order of 50ms or less. To achieve this goal the project focuses on approaches to pose-estimation which strike a good balance between accuracy and speed. The markerless pipeline consists of 3 stages: the processing of raw camera input at 60 frames per second, the 2D pose estimates of subjects in each view, and the final assembly into a full 3D skeleton. To manage this computationally heavy task at the desired latency, the overall markerless pipeline leverages the massively parallel computation abilities of modern CPUs and GPUs, allowing us to optimize every stage of the computations involved.
Vicon X Dreamscape Immersive @ SIGGRAPH 2023 : "The Clockwork Forest" Markerless VR Experience
The Markerless Mocap project was performed in collaboration with Vicon, the world leader mocap provider. First results of this joint collaboration were presented at SIGGRAPH 2023 on their exhibition booth, showcasing a six person, markerless and multi-modal real-time solve, set against Dreamscape’s LBVR adventure called The Clockwork Forest. With this showcase, Vicon earned a CGW Silver Edge Award for technological innovation.
Contributors
The following have contributed to this page
