Marker-tracking and deep-learning-based pose estimation for automatic crane work − Study on applying computer vision to construction sites

In response to the labor shortages caused by an aging society, many attempts have been made to improve productivity using information technology, such as sensors and automated heavy machinery. Crane work is one of the labor-intensive tasks in construction sites; hence, automation is expected. This study reports on a computer vision-based application for supporting crane work, specifically item-hoisting work. A pose estimation method is developed using marker-tracking and deep-learning-based global feature-tracking, which calculates position correction data for a crane and manipulation for a gyro instrument that controls the turn of a hoisted item. © 2021 Architectural Institute of Japan. All rights reserved.