Click here to enter text. Accurate tracking of material movement and processing paths throughout the manufacturing process is crucial for conducting rapid root cause analysis of quality issues and ensuring transparency in production history. Real-time location systems (RTLS) are widely used for this purpose, but high costs and complex installation environments limit their practicality. Computer vision-based object tracking is emerging as an alternative to overcome these limitations. Still, the temporary blockage of the field of view reduces the accuracy of Re-Identification (Re-ID) due to appearance similarity. The single-camera-based multi-object tracking (MOT) method is difficult to track continuously in a large space. The same object is recognized with different IDs even in a multi-camera environment. To solve these problems, we propose a grid-based multi-camera tracking technique that divides the overlapping camera field of view into a grid and matches IDs based on the object’s location information. The proposed method ensures that the continuity detected by each camera is preserved. Experiments conducted in simulated environments, including real-world occlusion scenarios, demonstrate that the proposed method outperforms the existing single-camera-based SORT tracking method in terms of identity matching.
Seo et al. (Thu,) studied this question.