相似性(几何)
人工智能
直觉
计算机科学
欧几里德距离
感知
运动(物理)
模式识别(心理学)
计算机视觉
心理学
图像(数学)
神经科学
认知科学
作者
Tatsuya Harada,So Taoka,Taketoshi Mori,Takehiro Sato
标识
DOI:10.1109/ichr.2004.1442140
摘要
In the imitation research, "how much similar two motions are" is an essential problem. This paper tries to clarify the criterion by which human evaluates the similarity of poses and motions. First we proposed the quantitative pose similarity evaluation which follows human's intuition. This proposed evaluation uses the optimized parameters such as attention to each human's region and the nonlinearity of the distance perception from questionnaires of similarities for many pairs of human poses. Secondly by comparing the conventional similarity evaluation, we confirmed the validity of the quantitative pose similarity evaluation among different pairs of human poses. Lastly the quantitative pose similarity evaluation was applied to the motion similarity evaluation. Verification of that appropriateness revealed that the quantitative pose similarity evaluation is also useful for the motion similarity evaluation. Since the quantitative pose similarity evaluation reflects the human's perception, this evaluation clarifies the criterions by which human evaluates the pose or motion similarity. The clarified criterions for pose or motion similarity evaluation are as follows: 1) the distance perception is processed in three dimensional positions, 2) the relationship between the similarity and the sum of the Euclidean distances for each region is nonlinear, 3) human fixates poses symmetrically, 4) fingertips are paid particular attention, 5) human pays more attention to further regions from the hip in the upper half of the body, and 6) human pays more attention to nearer regions from the hip in the lower half of the body.
科研通智能强力驱动
Strongly Powered by AbleSci AI