Many computer vision tasks rely on precise object localization and refining bounding boxes. Complex situations, including object rotations and varied scales, are typically too much for traditional approaches to handle. To update bounding boxes and perform orientation-aware localization, this research suggests an adaptive Particle Swarm Optimization (PSO) algorithm. During optimisation, the method takes orientation into account while representing particles and uses adaptive processes to tweak inertia weight and velocity updates. We develop a new fitness function that considers aspect ratio, orientation, and overlap for efficient evaluation. In comparison to conventional methods, the suggested strategy greatly enhances localization accuracy and robustness, according to the experimental results. This is particularly evident when objects undergo rotation or scaling. Based on these results, adaptive PSO seems like a promising tool for improving computer vision tasks, like object detection and localization.

Keywords: Object Localization, Bounding Box Refinement, Particle Swarm Optimization (PSO), Orientation Estimation, Adaptive Algorithms
View Full Paper