A Gait Sequence Analysis for IP Camera Using a Modified LBP

Abstract

This work proposes a method that is a modified form of the local binary pattern technique by changing the sorting method for the local binary pattern according to the blend direction to develop a new method, which is termed the Modified Local Binary Pattern (MLBP). After synchronizing and calibrating the gait sequence images, a cycle of images from the gait sequence is captured to form a gait energy image. The MLBP is then applied to the gait energy image to derive different blend direction images and to calculate the ability to recognize each blend direction image for feature selection. To allow classification, the Euclidean distance and nearest neighbor approaches are used. In the experiments that are carried out on the gait database, this approach achieves the best recognition rate, even when the appearance of the objects changes. Because the complexity is low, it is suitable for implementation in embedded systems and supports many Internet Protocol (IP) camera applications.