The Research on Multi-Vehicle Collaborative Navigation Method Based on SFM for Visual Sensor's Out-of-Field of View

In complex urban roadways and indoor environments, autonomous vehicles often encounter situations where their line of sight is obstructed. This is particularly true during turning maneuvers, where some autonomous vehicles may fall outside the observation range of another vehicle's visual sensors. This implies that traditional visual sensors are unable to directly obtain the relative positional relationship between two autonomous vehicles. Therefore, alternative reference methods must be relied upon to calculate the transformation matrix between the two vehicles, which can subsequently be used to estimate their motion trajectories. This paper proposes a novel method for calculating the transformation matrix between two autonomous vehicles in scenarios with limited visibility. The method is based on the Structure from Motion (SFM) algorithm, which involves feature extraction and the computation of the rotation matrix and translation vector to match common feature points in images captured by the two vehicles. Using these matched feature points, the transformation matrix between the two vehicles is calculated. Furthermore, to address the issue of insufficient common feature points caused by viewpoint differences in the images captured by the two vehicles, the paper introduces a strategy of retaining highly reliable image frames to effectively mitigate this problem. Experimental results demonstrate that the proposed algorithm can successfully calculate the transformation matrix between two autonomous vehicles under limited visibility conditions and use this matrix to accurately estimate the vehicles' motion trajectories.