Simultaneous Super Resolution and Moving Object Detection from Low Resolution Surveillance Videos

Video surveillance has become a significant method to preserve public security specifically in the forensic field where high resolution videos are vital. However, the security video footage captured is often found to be low in clarity and deteriorated with noise, blurs and uneven illumination which can cause false detections and wrong interpretations. In light of this relevant fact, this work proposes a novel technique that does video super resolution with simultaneous detection of the moving objects using a tensor based convex optimization problem. The ideas of multilinear algebra are utilized in the super resolution framework to implement re-weighted tensor nuclear norm minimization for separating the highly correlated scene background. The spatio-temporal continuity featured by the sparse foreground components is grabbed by the Tensor Total Variation (TTV) regularization to achieve precise object detection along with l1 regularization. The brilliance of the proposed approach is in the joint formulation and implementation of video super resolution and simultaneous moving object detection from the available deteriorated videos. This is solved using the Augmented Lagrangian Method (ALM) with alternating direction scheme and the results validate the superior performance of the proposed method against the compared approaches.