Self-Adaptive Multi-Sensor Framework for Precise Indoor Human Pose Estimation Based on LiDAR and mmWave Fusion

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Self-Adaptive Multi-Sensor Framework for Precise Indoor Human Pose 
Estimation Based on LiDAR and mmWave Fusion"

by

LIEU Kaixuan Ryan

Abstract:

Indoor human pose estimation (HPE) using single-modality LiDAR or mmWave 
sensors suffers from inherent limitations, including LiDAR’s background 
clutter and mmWave’s spatial sparsity. This project designs a robust 
multi-modal HPE framework integrating LiDAR and mmWave, comprising four core 
components: a mmWave-driven localization module, ROI-guided background 
reduction, k-NN-based mmWave-to-LiDAR feature transfer, and an adaptive 
self-updating background model. Validated on the MM-Fi dataset, the framework 
outperforms single-modality baselines and the state-of-the-art X-Fi 
framework, achieving 48.9% and 39.5% relative improvements in MPJPE and 
PA-MPJPE, respectively.

Date            : 28 April 2026 (Tuesday)

Time            : 16:00 - 16:40

Venue           : Room 2132A (near Lift 19), HKUST

Advisor         : Prof. CHAN Gary Shueng-Han

2nd Reader      : Dr. CHEN Hao