Human Skeleton Proportions Recovery from Monocular Data

This paper introduces a novel method for recovering the skeleton proportions for a human figure from monocular data. A perspective camera model is defined semi-automatically based on the user’s judgment. In the proposed method, key frames are first extracted from the source data automatically. A human skeleton model is then constructed to match all key frames under the established camera. An advantage of the proposed method is that no human posture validation is needed during the modeling process. The proposed method is tested to propose satisfactory results for some input data. The human model with recovered proportions can be used in further research involving body reconstruction or human motion reconstruction. Key-Words: Modeling, human figure, monocular