ARCHIVES

Abstract: In computer vision, object tracking in videos is an essential problem with implications ranging from autonomous driving to espionage. The methods, benefits, and drawbacks of object tracking approaches based on image processing are thoroughly examined in this work. A variety of traditional and modern techniques are covered in the paper, such as feature-enabled, model-enabled, contour-enabled, point-enabled, kernel-enabled, and deep learning-enabled tracking. The effectiveness of point tracking approaches in simulating linear and non-linear motion, including the particle filter and Kalman filter, is explored. The resilience of kernel-based tracking, especially the mean-shift technique, in real-time applications is analysed. The accuracy of contour-based tracking techniques, such as active contours, in detecting boundaries is examined. The paper also delves into model-based tracking, which utilizes 3D models to enhance tracking accuracy, and hybrid methods that combine the strengths of multiple approaches. Additionally, the advent of deep learning has introduced powerful methods which have significantly improved tracking performance through advanced feature representation and temporal dependency modelling. This work provides a concise yet comprehensive overview of the paper's content, outlining the various object tracking techniques covered and emphasizing their significance and applications.