Real-time object recognition and modeling for heavy-equipment operation

Abstract

Recognition of free-form objects located in environments that are difficult to characterize or are constantly changing is critical in providing interactive background information for construction worksite modeling. It also allows for accurate, efficient, and autonomous operation of heavy equipment in a broad range of construction tasks. This paper presents a real-time process for 3D modeling of a construction worksite scene and focuses on modeling of the target object via flash LADAR, which could be applied to autonomous heavy-equipment operation. The proposed method consists of three steps: noise reduction, object extraction, and 3D model generation. The whole process is fully automatic and is performed in nearly real time. The method was validated in field experiments with actual construction objects. The results show that the proposed method effectively recognizes construction objects, which could be used to enhance efficiency and productivity in the autonomous operation of heavy equipment.