Multi-Drone Collaborative Trajectory Optimization for Large-Scale Aerial 3D Scanning

Reconstruction and mapping of outdoor urban environment are critical to a large variety of applications, ranging from large-scale city-level 3D content creation for augmented and virtual reality to the digital twin construction of smart cities and automatic driving. The construction of large-scale city-level 3D model will become another important medium after images and videos. We propose an autonomous approach to reconstruct the voxel model of the scene in real-time, and estimate the best set of viewing angles according to the precision requirement. These task views are assigned to the drones based on Optimal Mass Transport (OMT) optimization. In this process, the multi-level pipelining in the chip design method is applied to accelerate the parallelism between exploration and data acquisition. Our method includes: (1) real-time perception and reconstruction of scene voxel model and obstacle avoidance; (2) determining the best observation and viewing angles of scene geometry through global and local optimization; (3) assigning the task views to the drones and planning path based on the OMT optimization, and iterating continuously according to new exploration results; (4) expediting exploration and data acquisition in parallel through multi-stage pipeline to improve efficiency. Our method can schedule routes for drones according to the scene and its optimal acquisition perspective in real-time, which avoids the model void and lack of accuracy caused by traditional aerial 3D scanning using routes of cultivating land regardless of the object, and lays a solid foundation for the 3D real-life model to directly become the available 3D data source for AR and VR. We evaluate the effectiveness of our method by collecting several groups of large-scale city-level data. Facts have proved that the accuracy and efficiency of reconstruction have been greatly improved.