Robust Composite Tracker of Objects in Video Images with Several Features and Reliable Design using Hierarchical Convolution Features

Abstract

Backdrop modeling is crucial in machine vision and image processing, such as automatic video surveillance and human-machine interaction (HCI). It involves modeling the background, eliminating pictures from the background image, and removing shadows from the final image. Moving objects leave a shadow behind them in the picture, which is represented as a backdrop image due to ambient noise and variations in brightness. Shadow motion detection is essential for recognizing objects in video streams, as shadow points are often misclassified as object points, resulting in segmentation and tracking problems. Many techniques for modeling shadows have been developed, but algorithms continue to be tested under various illumination settings. Block-based models have lately been adopted due to these circumstances, splitting the picture into equal blocks and determining movement by a collection of blocks in the most recent frames. This work presents a novel model based on PCA + LDA and hierarchical convolution features to overcome the difficulties in block-based techniques. The proposed technique considers the distinction between fixed and moving items based on the type of peripheral object.