Abstract
This study investigated human recognition of human-body postures. We used computer graphics to make natural and impossible (unnatural) poses of three-dimensional human-body models, and made two-dimensional images of them from different viewpoints to test recognition performance in a serial matching task. We found that the recognition performance was better for the natural poses than the impossible poses. Recognition performance with different views was worse than for identical views, indicating a view-dependent performance in human-pose recognition. The view-dependency was less for the natural poses than for the impossible poses. Recognition of the inverted poses was more difficult than of the upright poses, but there was no effect of inversion on the advantage of the natural poses. The advantage in recognition of natural postures suggests that the biomechanical constraints are utilized in human body recognition.
