Moving object detection in gigapixel-level videos using manifold sparse representation

Moving object detection (MOD) is one of the most important and challenging tasks in analyzing videos. Recently, emerging gigapixel-level videos have attracted considerable attention due to its large field of view and high spatial resolution. So far, there is not much research on moving object detection in the field of gigapixel-level videos. To detect moving objects in gigapixel-level videos, we propose a novel manifold sparse representation (MSR) method for detecting moving objects in gigapixel-level videos. The innovation of this method lies in introducing sparse representation to promote global structure and utilizing manifold learning to retain the local structure of moving objects. Then, to solve the explicit solutions of this problem, an efficient optimization scheme based on the manifold alternating direction method of multipliers (MADMM) is developed. Finally, the extensive experiments on three challenging benchmark MOD datasets, three samples from the gigapixel-level PANDA dataset (gigaPixel-level humAN-centric viDeo dAtaset), and one real wide area video dataset are executed to demonstrate the superiority of the proposed MSR. In particular, the detection performance on PANDA is improved by an average of 5%, which suggests that the proposed method is promising for gigapixel-level videos.