AN EFFECTIVE DIRECTIONAL MOTION DATABASE ORGANIZATION FOR HUMAN MOTION RECOGNITION

Automatic recognition of human motions has an increasing demand in the recent visionary world. However, with the registration of large number of motions from varying viewpoints, the necessity for an effective motion database for recognition has be- come a vital issue. In the context of motion database development, this paper proposes a directional database organization for human motion recognition. This organization parti- tions the motion database into several sub-databases on the basis of camera orientation. Separate feature spaces are constructed, and correspondingly directional sub-databases are built, leading to the constitution of the complete motion database. The directionally similar but semantically different motions are properly distinguished. To show the ro- bustness of the proposed organization for recognizing human motions, a set of motions captured from varying viewpoints is analyzed. An eigenspace representation is employed as a generic feature space that sufficiently characterizes the motion features. Motion His- tory Image (MHI) and Exclusive-OR (XOR) image representations are used as motion templates where MHI is found performing better than XOR image. The experimental re- sults show high-level of satisfactory performance and claim the signicant improvement over earlier developed systems.