From Individuals to Crews, AI Brings Teamwork Into Construction Productivity Analysis

Why It Matters

By capturing crew dynamics, the technology provides more accurate productivity metrics and supports smarter, collaborative construction automation.

Key Takeaways

• AI recognizes individual and crew activities from site videos
• Multi‑granular framework links actions, crews, site operations
• Achieves >73% F1 score on real construction footage
• Crew analysis yields more accurate productivity insights
• Future work adds temporal modeling for dynamic interactions

Pulse Analysis

Construction productivity has long relied on manual observations and work‑sampling techniques that are costly, subjective, and difficult to scale. Recent advances in computer vision have enabled the detection of single‑worker actions, yet they fall short of representing the collaborative nature of most on‑site tasks. Understanding how teams coordinate is essential for accurate performance measurement, safety oversight, and the integration of automated equipment.

The new AI system from National Taiwan University bridges this gap by modeling workers as graph nodes and learning their spatial relationships to infer crew‑level activities. Trained on extensive video data from Taipei projects—including rebar placement, formwork installation, and concrete pouring—the framework delivers an F1 score exceeding 73%, demonstrating reliable recognition of both individual and collective actions. By distinguishing value‑added work from idle periods at the crew level, the technology offers actionable insights that surpass traditional individual‑focused analytics.

Industry implications are significant: real‑time crew monitoring can streamline project scheduling, reduce non‑productive time, and enhance safety protocols by flagging hazardous group behaviors. Moreover, the ability to quantify teamwork lays a foundation for effective human‑robot collaboration, where robots must adapt to dynamic human group patterns. Future research aims to incorporate temporal modeling, expanding the system's capacity to track evolving interactions and broader task sets, ultimately driving smarter, more adaptive construction ecosystems.