APPLICATION OF HYBRID MODELS FOR BACKGROUND SUBTRACTION IN VIDEOS

Abstract

In recent years, Background Subtraction (BGS) has become a major topic of study in the field of Computer Vision for performing moving object detection. BGS is typically utilized in video surveillance cameras to monitor moving objects and is a widely used technique in the field of computer vision for non-stationary object identification and tracking. Hybrid models are one of the many types of approaches that can be found in the BGS literature as a result of extensive ongoing studies. Hybrid models are created by combining two or more models, allowing them to benefit from each other's strengths while overcoming the weaknesses of the original models. In this report, firstly some of the recently developed hybrid models like Hierarchical Modeling and Alternating Optimization (HMAO), randomized dynamic mode decomposition (rDMD), Adaptive Motion Estimation and Sequential Outline Separation (AME+SOS) etc. are descriptively analyzed using a tabular form of review for a clear and easy understanding in addition with the comparative analysis which is performed based on F-m values of the models for a video sequence from the very popular CDnet dataset. Furthermore, in this project a new method based on Hybrid modelling technique that combines two models, namely Robust Principal Component Analysis (RPCA) using Principal Component Pursuit (PCP) and randomized Singular Value Decomposition (rSVD) is proposed. Video sequence is decomposed into frames and a data matrix is created with frames as a column vector and the above-mentioned methods are then applied on the data matrix to achieve the backdrop and foreground of the video. Experiments are run on two well-known datasets, CDnet (2014) and BMC, as well as a random YouTube video sequence. In comparison to the classic PCP model, experimental results reveal that the proposed model has lower computing complexity and produces results in less time.