arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1503.08223cs.CV
收藏

观察到的人的形状、姿势和动作的识别和解释的系统观

A System View of the Recognition and Interpretation of Observed Human Shape, Pose and Action

Montana State University-Bozeman(蒙大拿州立大学波兹曼分校);General Intelligence Corporation(通用智能公司)

David W. Arathorn

摘要:

有生理学证据表明,我们从2D视觉图像(双眼或单眼)解释人类姿势和动作的能力涉及运动皮层的电路以及大脑的视觉区域。这意味着,运动皮层解决逆运动学的能力是足够灵活的,适用于运动规划以及作为一个生成模型的视觉处理的人的数字,尽管不同的功能要求的两个任务。本文提供了一个计算模型的视觉和运动区域之间的合作:换句话说,一个重要的一类大脑计算的系统视图。该模型统一的任务,视觉变换发现,逆运动学,并适应形态变化,使用地图搜索电路算法的几个实例中所涉及的单独的逆问题的解决方案。虽然该文件是加权对神经生物学假设的阐述,从数学形式化的问题,神经元电路,算法表达的解决方案也是一个功能机器视觉系统的人的数字识别,和3D姿态和身体形态重建从单眼,透视输入图像。利用能够施加各种内源性和外源性约束的逆运动生成模型,机器视觉实现获得了此类系统中当前唯一的特征。

英文摘要:

There is physiological evidence that our ability to interpret human pose and action from 2D visual imagery (binocular or monocular) engages the circuitry of the motor cortices as well as the visual areas of the brain. This implies that the capability of the motor cortices to solve inverse kinematics is flexible enough to apply to both motion planning as well as serving as a generative model for the visual processing of human figures, despite the differing functional requirements of the two tasks. This paper provides a computational model of the cooperation between visual and motor areas: in other words, a system view of an important class of brain computations. The model unifies the solution of the separate inverse problems involved in the task, visual transformation discovery, inverse kinematics, and adaptation to morphology variations, using several instances of the Map-seeking Circuit algorithm. While the paper is weighted toward the exposition of a neurobiological hypothesis, from mathematical formalization of the problem to neuronal circuitry, the algorithmic expression of the solution is also a functional machine vision system for human figure recognition, and 3D pose and body morphology reconstruction from monocular, perspective-less input imagery. With an inverse kinematic generative model capable of imposing a variety of endogenous and exogenous constraints the machine vision implementation acquires characteristics currently unique among such systems.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑