UAV-Based Multi-Sensor Data Fusion for 3D Building Detection

Three-dimensional building extraction is crucial for urban planning, environmental analysis, and autonomous navigation. One method for data collection involves using unmanned aerial vehicles (UAVs), which allow for flexible and rapid data acquisition. However, accurate 3D building extraction from these data remains challenging due to the abundance of information in high-resolution datasets. To tackle this problem, a novel UAV-based multi-sensor data fusion model is developed, which utilizes deep neural networks (DNNs) to enhance point cloud segmentation. Urban datasets, acquired by a UAV equipped with a Zenmuse L1 payload, are collected and used to train, validate, and test the DNNs. It is shown that most building extraction results have precision, accuracy, and F-score values greater than 0.96.