Event camera-based human pose estimation via hybrid spiking-point cloud neural architecture

Event camera-based human pose estimation faces fundamental trade-offs between temporal precision and computational efficiency. We propose a novel hybrid spiking-point cloud network that overcomes these limitations through a three-modal channel splitting strategy. The system incorporates cross-modal...

Full description

Saved in:
Bibliographic Details
Main Authors: Sichao Tang, Hengyi Lv, Xiangzhi Li, Yuchen Zhao, Yisa Zhang, Yang Feng
Format: Article
Language:English
Published: Elsevier 2025-09-01
Series:Results in Engineering
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2590123025028269
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Event camera-based human pose estimation faces fundamental trade-offs between temporal precision and computational efficiency. We propose a novel hybrid spiking-point cloud network that overcomes these limitations through a three-modal channel splitting strategy. The system incorporates cross-modal adaptive fusion and asynchronous skeleton constraints, achieving superior accuracy while maintaining efficiency in complex scenes and rapid motion scenarios. The system also incorporates a cross-modal adaptive fusion mechanism that dynamically adjusts weights across different modalities, and an asynchronous skeleton constraint module that leverages human anatomical prior knowledge to constrain prediction results. Compared to existing methods, our network architecture more effectively processes the sparse asynchronous characteristics of event data, achieving a better balance between accuracy and efficiency, particularly excelling in complex scenes and rapid motion scenarios. Code will be available at: https://github.com/DVSexplorer/EHPE-via-HSPC.
ISSN:2590-1230