A Globally Referenced and High-precision Visual Positioning Device Based on Multi-sensor Fusion Technology

Abstract

Megatrends like autonomous driving and mapping with unmanned aerial vehicles (UAV) are pushing the boundaries for new challenges and opportunities in reality capture applications. In the last years, dramatic advances in computer vision and artificial intelligence have opened up new perspectives in navigation and positioning systems. Such systems are typically based on sophisticated sensor fusion algorithms integrating data from multiple navigation sensors. Numerous aerial and land-based applications that require high-precision position and attitude information in real time rely upon an accurate and robust determination of the absolute six degrees of freedom (6DoF - x, y, z, roll, pitch, heading), especially by combining GNSS and inertial measurement units (IMU). The major benefit of including GNSS is to obtain position information directly in a global reference frame. Additional sensor systems such as camera-based visual inertial systems (VIS), LiDAR, and scanning will benefit from the precise attitude information and globally referenced positions, and vice versa. This paper aims at analyzing the potential and performance of combining simultaneous image capturing and absolute 6DoF information for high-precision positioning application. The 3D camera pose estimation by means of a GNSS-IMU system is combined with images and SLAM algorithms. This visual positioning approach is innovative and enables accurate remote points measurement by fusing GNSS-IMU with terrestrial photogrammetry. As a result, globally referenced visual positioning with centimeter-level accuracy becomes possible even in GNSS-denied environments. Land-based applications such as navigation, positioning and scene documentation can be performed conveniently in a global reference frame. In this study, global pose information, camera captures, and computer vision algorithms are used to measure points in images. The results are compared against a reference field with a higher order of accuracy to demonstrate the performance of the proposed visual positioning approach. Representative tests were carried out by considering various camera-to-object distances and trajectories varying in length and geometry to evaluate the accuracy and reliability of this novel sensor fusion technology. The tests show that the combination of visual inertial positioning with a high-precision 6DoF system offers enormous potential in land-based applications where globally referenced data is demanded within the shortest time. By bringing together sensor fusion and computer vision, significant improvements can be achieved for more productive reality capture capabilities with a high level of accuracy and reliability.