Enhanced object tracking with artificial bee colony, motion modeling, and deep learning

Abstract

As a fundamental aspect of computer vision, visual object tracking supports a wide array of applications, notably in transport infrastructure and advanced industrial automation. Although correlation filter-based trackers demonstrate robust performance, they face persistent limitations including scale changes, object occlusion, boundary artifacts, and complex background interference. To address these issues, we have introduced an approach that combines artificial bee colony (ABC) optimization, deep neural architectures, and Kalman filtering techniques. Our methodology begins with reliability assessment of the tracking pipeline, proceeding to compute target confidence measures at the predicted position, followed by an adaptive update mechanism. The proposed system leverages ABC optimization for dynamic scale adaptation while employing Kalman filtering to model inter-frame target motion dynamics. Comprehensive evaluation across multiple benchmark datasets demonstrates our method's efficacy, precision, and resilience, achieving enhanced performance relative to existing state-of-the art approaches.