Detection of pedestrians in motion with rotation and scale variation

In this paper, a new framework for tracking non-rigid objects is introduced. By spatially masking the target with an isotropic kernel, a spatially smooth similarity function can be defined, and the target localization problem is then reduced to a search in the basin of attraction of this function. The smoothness of the similarity function allows for the application of a gradient optimization method by adopting a background weighted histogram. The proposed approach is effective for detecting objects in complex background and occlusion. Experimental results using various test videos containing objects in motion with rotation and scale changes show us superb detection accuracy over the conventional method.