Detection of Shadow Changes and Object Movement in Colour Images of Outdoor Scenes

Abstract

Detecting object movement in real-world environments can present a challenge in image processing applications. Variations in illumination, the distribution of shadows, and atmospheric scatter introduce uncertainties, and potential confounds. To accurately detect object movement across successive images, these uncertainties must be effectively managed. Current methods for object movement and shadow change detection have notable limitations. Supervised techniques require large annotated datasets, making them resource-intensive, while unsupervised methods often use computationally expensive approaches that utilise image patches. Although shadow detection methods can perform well on small image patches, they often struggle with real-world complexity. Furthermore, many methods rely on continuous image sequences or assumptions about background and foreground decomposition, limiting their generalisation across diverse environments. The proposed algorithms with local and global matching address these complexities by being both training-free and assumption-free, requiring only a pair of images. Local matching simulates human comparison of specific regions to detect differences, while global matching reflects broader contextual adaptation. Local matching involves segmenting regions, forming composites, and labelling regions to detect shadow changes. Global matching, similar to the way the human visual system adjusts to its surroundings, focuses on broader contextual alignment by adapting image domains, segmenting shadow changes, and isolating object movement from shadows. These algorithms demonstrate state-of-the-art performance across diverse indoor and outdoor scenes.

Date of Award	14 Apr 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	David Foster (Main Supervisor) & Hujun Yin (Co Supervisor)