PROJECT TITLE :

An Empirical Review of Deep Learning Frameworks for Change Detection Model Design, Experimental Frameworks, Challenges and Research Needs

ABSTRACT:

One of the fundamental objectives of computer vision and video analytics is the separation of video frames into distinct foreground and background regions. Visual change detection is one of the fundamental tasks in these two fields. Anomaly detection, object tracking, traffic monitoring, human machine interaction, behavior analysis, action recognition, and visual surveillance are some of the applications of change detection. Background fluctuations, illumination variation, changes in the weather, intermittent object motion, shadows, fast/slow object motion, camera motion, heterogeneous object shapes, and real-time processing are some of the challenges that come with change detection. Historically, this issue has been addressed and resolved through the application of hand-crafted features and background modeling strategies. Over the past few years, successful adoption of deep learning frameworks has occurred for the purpose of robust change detection. An empirical analysis of the most cutting-edge applications of deep learning to the problem of change detection is the goal of this article. To be more specific, we present a comprehensive analysis of the technical aspects of a variety of model designs and experimental frameworks. We provide a model design-based categorization of the existing approaches, which includes the 2D-CNN, the 3D-CNN, the ConvLSTM, multi-scale features, residual connections, autoencoders, and GAN-based methods. Additionally, an empirical analysis of the evaluation settings utilized by the currently available deep learning methods is provided in this article. This is an initial attempt, to the best of our knowledge, to conduct a comparative analysis of the various evaluation frameworks that are used in the various deep change detection methods that are currently available. In the end, we discuss the areas that require further research, outline some potential future directions, and offer our own conclusions.