Vision-Based Activity Analysis Framework Considering Interactive Operation of Construction Equipment

摘要

Automated activity analysis is vital for efficient productivity management on construction sites. A vision-based method has received attention due to its efficiency in identification and tracking. Although existing vision studies showed applicability, they did not consider interactive operations between equipment, and thus deteriorated analysis performance. To tackle the limitation, this paper proposes a vision-based activity analysis framework considering interactive operations of construction equipment. The framework includes four main processes: detection, tracking, individual action recognition, and interaction analysis. For the feasibility analysis, the individual actions of excavators were analyzed automatically using tracking-learning-detection and bags-of-features. This framework was validated with video images collected from earthmoving construction sites, and the average precisions of detection, tracking and action recognition were 88.0%, 88.0%, and 83.6% respectively. The experimental results showed the developed approach was able to identify independent actions of equipment, but the interactive operations should be considered to be more reliable activity analysis.