Tool storage systems are an integral component of the production chain in modern manufacturing facilities. Automated vertical storage systems are commonly employed to store and manage tools and equipment required for rapid replacement or re-tooling during the production process. In such a scenario, any error made by a warehouse operator can disrupt the inventory system, leading to operational issues or even halting the production line. To address the challenges of storage control and operator error identification, this paper proposes a vision-based system capable of detecting changes within the storage space and determining their directionality. The proposed solution leverages a custom synthetic dataset generation process and a hybrid processing method, combining a 6-channel enhanced YOLOv8 (You Only Look Once) model with Structural Similarity Index Measure (SSIM) analysis. This approach effectively identifies the location and direction of changes (e.g. object removal or addition) and is characterised by robustness to domain shifts and other disturbances, such as variations in illumination or object relocation, which commonly occur during normal operation. The enhanced model utilises a 6-channel input, integrating ”before” and ”after” images while retaining full colour space information - a capability not achievable with the standard YOLO models. Furthermore, the two-stage processing method that incorporates SSIM analysis significantly improves the recall rate of the developed solution. Comprehensive validation on prepared test datasets demonstrated an F1-score of 95.1, with Average Precision (AP50) and Average Recall (AR50) of 88.1 and 79.7, respectively.