Machine Learning–Based Online Visual Tracking with Multi-featured Adaptive Kernel Correlation Filter

Abstract

An important challenge in computer vision is following moving objects as they travel throughout video sequences. Object tracking and detection are crucial tasks. This is the basis for many higher-level automated applications in many fields, including augmented reality, surveillance and motion capture for moving object recognition. We proposed the use of scale adaptive kernel support-correlation filter (SKSCF) algorithm. The basis for the application of IVS in this work is due to its active alternating optimisation approach for visual tracking, from which we build an analogous formulation of support vector machine (SVM) model along with the circulant matrix expression. The proposed work was designed to achieve the following: construct a video sequence for tracking moving objects; plan an experimental setup for moving object identification; and develop and build a moving object tracking algorithm, which was applied to a recorded video sequence. The background subtracting is a technique for detecting objects in video pictures captured by a single lens over a static background, that requires that the camera be stationary. Various videos with a fixed camera with single and multiple objects are tested. Motion-based algorithms for locating and observing a specific moving item of interest can be developed. The frame and feature of the object have been identified using SCF feature extraction to match the desired object. An extensive testing on the OTB-50 standard has revealed that the suggested work produced the best results admirably when compared to numerous cutting-edge monitors, getting an AUC score of 90.6%.