Real-time structural crack detection in buildings using YOLOv3 and autonomous unmanned aerial systems

An innovative approach for efficient and controlled drone-based building inspection has been introduced in this work. By integrating hardware and simulation, the system ensures efficient testing and validation using jMAVSim. Unlike conventional techniques that rely on onboard computing devices for real-time object detection and continuous wireless data transmission, posing bandwidth challenges and limited user control over the drone’s actions during detection, our approach utilizes long-range data transmission, eliminating onboard computing needs. The system establishes a networked drone damage detection system (DDS), offering real-time outputs and user control for efficient structural inspection, making it an efficient autonomous solution for structural inspection. The proposed system utilizes a ground station (Laptop) as a hardware platform, integrating YOLO-v3 for object detection, severity classification and action generation based on live video streamed from a drone’s long-range transmitter. The system’s efficacy is assessed using a dataset featuring diverse damage types. During a survey, if damage is detected, the ground station system employs severity classification to trigger MAVLink commands, pausing the mission. Based on detected damage severity, action decisions are made and transmitted to the drone through telemetry. Test outcomes demonstrate the ground station system’s capability to detect cracks and appropriately respond such as halting a survey when structural damage is identified. The proposed method achieves a mean average precision (mAP) of 74.67%, processing 9 to 11 frames per second with a batch size 34. This innovative approach optimizes building inspections by leveraging drone technology, offering enhanced precision, reduced risk and streamlined operations.