一种面向大场景视频的动态三维人群重建方法

• Format: Patent
• Language: Chinese
• Published: 18.04.2025
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

本发明公开了一种面向大场景视频的动态三维人群重建方法，涉及三维视觉技术领域。本发明与之前的大场景静态多人重建方法相比，解决了由于动态遮挡而导致的姿态特征损失的问题；基于自上而下的多人重建框架，通过迭代优化的方式得到鲁棒的人群三维位置、姿态和形状，设计人群分组优化范式来聚类具有相似运动轨迹的个体，并通过人物运动的未遮挡分数使用未遮挡序列自适应修复遮挡序列，还设计了异步运动一致性损失进行联合优化，利用时间规整技术找到姿势序列之间的最佳对齐，解决动态遮挡带来的运动丢失问题；通过使用变分自编码器训练了人体运动先验模型，有效增强了运动的真实感和平滑度。 The invention discloses a large-scene video-oriented dynamic three-dimensional crowd reconstruction method, and relates to the technical field of three-dimensional vision. Compared with a previous large-scene static multi-person reconstruction method, the problem of attitude feature loss caused by dynamic shielding is solved; on the basis of a multi-person reconstruction framework from top to bottom, a robust three-dimensional position, posture and shape of a crowd are obtained through an iterative optimization mode, a crowd grouping optimization normal form is designed to cluster individuals with similar motion trails, and an unshielded sequence is used for adaptively repairing a shielded sequence through unshielded fractions of person mo