Vision‐based localization for UAV swarm traversal control: Theory and experiments

Abstract Vision‐based localization is essential for UAV swarm traversal control in fields such as military reconnaissance and disaster relief. However, it presents challenges including collision risks and the precise localization and tracking of target areas. Therefore, this paper proposes a vision‐based collision‐free formation control method for UAV swarm traversal missions. Initially, LSD (line segment detector) algorithm is employed for target area detection, and a precise localization model is constructed using pinhole projection and triangle similarity principles. Subsequently, to ensure safe formation flight, a repulsive force function model based on Hooke's law is derived, incorporating both the relative position and velocity of the UAVs. Based on this, a collision‐free formation controller is designed within the consensus theory framework to meet formation transformation requirements. Finally, the experiment is conducted with 10 UAVs, including two with cameras. They are divided into two groups, each with five UAVs. In each group, the camera‐equipped UAVs detected and localized the target area, transmitting the center position to the other UAVs. Subsequently, both groups traversed the target area in formation. The results show that the UAV swarm successfully traversed two 2.5 m*2.5 m boxes, and the minimum distance from the boxes was kept at 0.5712 m, and there was no collision between UAVs during formation changes and traversal.