Partial Body Image Restoration for Skeleton Estimation

The system performs preliminary action by restoring missing body parts in the image data before skeleton estimation is performed. The restoration unit completes partial body images using learned relationships from training data, ensuring that the skeleton estimation model receives complete body information. This preliminary restoration step enables accurate skeleton estimation without requiring manually prepared whole-body images during actual authentication operations.

The restoration unit acts as an intermediary between the input image data and the skeleton estimation model. It transforms partial body images into complete body representations by filling in missing regions based on learned patterns, thereby mediating the information gap between partial inputs and the requirements of the skeleton estimation algorithm.

The system creates virtual copies of missing body parts by learning from training data containing whole-body images. During the training phase, the system learns the relationships between visible and hidden body parts, enabling it to generate accurate restorations of missing regions. This copying approach allows the model to work with partial images during authentication while maintaining accuracy equivalent to whole-body image processing.

The training process performs preliminary learning of body structure relationships in advance. By training on whole-body images during the offline phase, the system acquires knowledge about how different body parts relate to each other, which is then applied during authentication to restore missing parts from partial images without requiring manual preparation of complete body images at runtime.