A Survey on Multi-Sensor 3D Human Pose Estimation

PhD Qualifying Examination


Title: "A Survey on Multi-Sensor 3D Human Pose Estimation"

by

Mr. Zhuoxuan PENG


Abstract:

3D human pose estimation aims to recover the spatial configuration of the 
human body from sensor observations. Although monocular RGB methods have 
advanced rapidly, depth ambiguity, occlusion, and crowded scenes still limit 
their reliability. Multi-sensor systems mitigate these limitations by 
combining observations across viewpoints and sensing modalities. The relevant 
literature, however, spans sensing setups with different assumptions, 
evaluation practices, and deployment constraints. This thesis presents a 
survey of multi-sensor 3D human pose estimation that brings multi-view and 
multi-modal approaches into a common framework. The survey organizes 
multi-view methods by how camera observations are integrated and multi-modal 
methods by the additional physical cues introduced beyond RGB. It also 
compares representative datasets and evaluation metrics, emphasizing 
multi-person support, calibration, synchronization, missing observations, 
computational cost, and privacy. The thesis concludes with directions for 
robust, practical, and privacy-aware multi-sensor 3D human pose estimation.


Date:                   Friday, 26 June 2026

Time:                   1:00pm - 3:00pm

Venue:                  Room 5501
                        Lifts 25/26

Committee Members:      Prof. Gary Chan (Supervisor)
                        Prof. Pedro Sander (Chairperson)
                        Dr. Dan Xu