Efficient CNN-based Object ID Association Model for Multiple Object Tracking

In the rapidly evolving field of machine learning, advancements in Multiple Object Tracking (MOT) are driven by the integration of sophisticated deep learning models. MOT plays a crucial role in enhancing the quality of multimedia experiences by establishing the groundwork for visual consistency, interactive applications, effective content organization, and a more immersive and captivating user experience across a variety of multimedia platforms. A significant challenge in Multi-Object Tracking (MOT) involves resolving the ID association problem, which entails accurately assigning and maintaining unique identifiers for objects across consecutive frames. Successfully managing this task is crucial for ensuring continuous and consistent tracking of individual objects, thereby enhancing the accuracy and reliability of MOT systems in diverse real-world scenarios. However, it is difficult due to issues like occlusions and variations in appearance. This research investigates the efficacy of a CNN-based approach in achieving accurate ID association in MOT, with a specific focus on the similarity between targets. A tailored Siamese Network is developed and assessed on the created dataset, revealing outstanding performance in terms of accuracy in tracking capabilities. The proposed model adeptly manages challenges such as low-light conditions and occlusions, surpassing state-of-the-art methodologies. The results affirm the suitability of the CNN-based model for ID association in MOT, underscoring its superior performance and heightened resilience, achieving an accuracy of 91.98%.