Determining the number and positions of people is critical throughout the entire lifecycle of construction projects for effective safety and productivity management. Although recent vision-based methods for people counting have advanced, most have uti...
Determining the number and positions of people is critical throughout the entire lifecycle of construction projects for effective safety and productivity management. Although recent vision-based methods for people counting have advanced, most have utilized single-camera and have been implemented at macro-levels, constraining their applicability in large areas due to the limited field of view (FOV) and increasing counting errors. To address these challenges, this research proposes a novel multi-camera analysis methodology for zone-level people counting that incorporates person re-identification (Re-ID). Re-ID is the process of matching the same individuals appearing in different views and is an essential technique for fusing information from multiple cameras. The research began by processing video images using Grounded-SAM2 to detect and track people, thereby accumulating bounding boxes and coordinates. Considering the multi-camera layouts, appropriate calibration-free camera network maps were constructed, and the resulting spatiotemporal cues together with the tracking results were fed into the Re-ID model. These Re-ID outputs were then passed to a counting module that produced zone-level people counting results over finely partitioned zones of the whole monitored space. Finally, the proposed methodology was evaluated using three video datasets collected from one building and two construction sites. As a result, it was confirmed that the proposed method can comprehensively derive and analyze zone-level people counting information in large spaces of construction projects from sites to buildings, and it can provide reliable counts even in regions with blind spots and frequent occlusions. In addition, it is expected that the framework can support efficient people management throughout the lifecycle of construction projects and, in turn, enhance both productivity and safety.