Integrating UAVs and AI for Residential Building Inspections

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Nov 2 2023

In a recent publication in the journal Buildings, researchers introduced an empirical case study that explores the integration of unmanned aerial vehicles (UAVs) and artificial intelligence (AI) for inspecting residential buildings.

Background

In recent years, the deterioration of buildings and infrastructure has become a significant concern. To address this issue, structural health monitoring and condition inspection have gained importance. Effective management of aging structures requires early defect identification and systematic maintenance to extend their lifespan.

Traditional periodic safety inspections have drawbacks, including subjectivity, time consumption, and high costs. The scarcity of inspection experts further complicates matters, emphasizing the need for more efficient and objective inspection methods. In response to these challenges, innovative visual inspection techniques that incorporate UAVs and AI have been explored.

While these methods have shown promise in bridge inspections, their application to residential buildings presents unique challenges, particularly concerning privacy concerns and varying building structures. The current study provides a detailed analysis of safety inspection cases conducted using UAV-AI-based methods.

UAV-AI-based inspection process for residential buildings

Researchers adopt an explanatory case analysis method to gain in-depth insights into the utilization of UAV and AI technologies in automatic building inspection. The process comprises four key steps: preliminary preparations, data acquisition, AI defect detection, and 3D reconstruction with defect extraction.

Preliminary Preparations: To facilitate UAV data collection for residential buildings, a set of preliminary measures is taken to ensure safe and effective flights. This includes on-site evaluations of aviation regulations, surveys of the surrounding area, adherence to national flight regulations, obtaining official approvals and permissions, and comprehensive site investigations.

Data Acquisition: Researchers proposed a comprehensive UAV flight plan for precise data acquisition. Two distinct approaches are used: automatic flight path planning and manual flight. Automatic flights collect exterior building data and environmental information, while manual flights provide reference points for 3D modeling. The data acquisition aims to minimize distortion and ensure high-quality data collection, covering architectural features such as eaves, windows, and corners.

AI for Defect Detection: This stage introduces the AI engine-based defect detection method, a crucial component of the inspection process. High-quality data is essential for training the AI model. Images capturing defects are selected and meticulously annotated, creating bounding boxes around defect areas. The study utilizes the Faster Region-Convolutional Neural Network (Faster-RCNN) model for defect detection, which excels in both defect recognition and precise localization.

3D Reconstruction and Defect Extraction: The images collected through the flight plans, both automatic and manual, are used to reconstruct 3D models of residential buildings. Irrelevant data is removed to expedite model synthesis. The 3D modeling process is fine-tuned manually for improved model quality. The defects detected by the AI model are incorporated into the 3D models, allowing for a visual representation of defects on the elevation map for each building.

Case Study Selection: The study analyzed five cases selected following preliminary investigations. These buildings, aged over 30 years, have been steadily deteriorating due to the lack of systematic government maintenance and management. Field tests were conducted using AI-equipped UAV inspection methods to assess the efficacy of the proposed approach across diverse building conditions, including factors such as ashlar lines, heights, and years of use.

Evaluation of the automatic defect detection method

Researchers evaluated the outcomes of automatic defect detection from surface images of building structures. Bounding boxes on these images were generated based on the intersection over union (IoU) metric, with a threshold of 50 percent. Precision-recall curves were used to assess detection performance. By using a pre-trained automatic fault detection model on inspected buildings in five different scenarios, the study verified that the model was applicable. In each case, about 40 photographs were annotated to test the model. However, the results indicated that the pre-trained model exhibited suboptimal performance, necessitating further fine-tuning.

3D Model-Based Defect-Location Extraction: To convey a building's state visually, 3D modeling of image data was employed. AI networks were used to detect the localization and type of defects, and an indexing method within the flight path plan tracked defect information on the building exteriors. This information allows inspectors to assess defect location and shape visually.

Lessons Learned and Limitations: The case analysis revealed key considerations for successfully implementing UAVs and AI techniques in building inspection and monitoring. Site investigations, flight planning, privacy concerns, and data quality were identified as crucial factors. The study proposes solutions to overcome the limitations and challenges posed by using UAVs and AI technologies in residential building inspection.

Conclusion

In summary, the current study explores the application of UAVs in inspecting and monitoring residential building defects integrated with AI techniques. Researchers propose a UAV and AI-integrated inspection process illustrated through a comprehensive case study. The analysis identifies limitations in current processes and highlights factors for effective AI-integrated UAV inspections.

Future work includes improving AI-based multi-defect detection models and addressing digital twin models for building information transformation, contributing to future AI-UAV-based building inspections and advancements.