My caption

Automated Wall-Climbing Robot for Concrete Construction Inspection

Publication
Journal of Field Robotics (JFR), 2022, DOI: 10.1002/rob.22119

Abstract: Human-made concrete structures require cutting-edge inspection tools to ensure the quality of the construction to meet the applicable building codes and to maintain the sustainability of the aging infrastructure. This paper introduces a wall-climbing robot for metric concrete inspection that can reach difficult-to-access locations with a close-up view for visual data collection and real-time flaws detection and localization. The wall-climbing robot is able to detect concrete surface flaws (i.e., cracks and spalls) and produce a defect-highlighted 3D model with extracted location clues and metric measurements. The system encompasses four modules, including a data collection module to capture RGB-D frames and inertial measurement unit data, a visual–inertial navigation system module to generate pose-coupled keyframes, a deep neural network module (namely InspectionNet) to classify each pixel into three classes (background, crack, and spall), and a semantic reconstruction module to integrate per-frame measurement into a global volumetric model with defects highlighted. We found that commercial RGB-D camera output depth is noisy with holes, and a Gussian-Bilateral filter for depth completion is introduced to inpaint the depth image. The method achieves the state-of-the-art depth completion accuracy even with large holes. Based on the semantic mesh, we introduce a coherent defect metric evaluation approach to compute the metric measurement of crack and spall area (e.g., length, width, area, and depth). Field experiments on a concrete bridge demonstrate that our wall-climbing robot is able to operate on a rough surface and can cross over shallow gaps. The robot is capable to detect and measure surface flaws under low illuminated environments and texture-less environments. Besides the robot system, we create the first publicly accessible concrete structure spalls and cracks data set that includes 820 labeled images and over 10,000 field-collected images and release it to the research community.

Related