A Local Feature-Based Approach For Activity 3-D Skeleton Posture Retrieval
In this work, a novel method is proposed that utilizes depth information of human body to retrieve appropriate
posture from a database. The retrieval method utilizes Enhanced Independent Component Analysis (EICA) of depth postures
of different classes over which Linear Discriminant Analysis (LDA) is performed. Then, the postures of each class are ranked
in sorted order based on the distance to their class center. To retrieve a testing posture from the training database, the posture is
projected on the feature space based on which its class is determined by measuring the distances to all class centers. Finally, a
binary search is done in the corresponding class to find the nearest distance of the corresponding posture and the one is chosen
with the least distance.
Keywords- EICA, LDA, Depth, Posture.