Sky-GVIO: an enhanced GNSS/INS/Vision navigation with FCN-based sky-segmentation in urban canyon

Overview

• Accurate and reliable positioning is essential for autonomous driving
• In urban environments, challenges like non-line-of-sight (NLOS) from buildings, trees, and structures can degrade positioning
• The paper proposes a sky-view image segmentation algorithm and a novel NLOS detection and mitigation (S-NDM) algorithm to address these challenges
• The system, called Sky-GVIO, combines Global Navigation Satellite Systems (GNSS), Inertial Measurement Units (IMU), and visual features for continuous and accurate positioning
• The system also harmonizes Single Point Positioning (SPP) and Real-Time Kinematic (RTK) approaches for versatility and resilience

Plain English Explanation

Autonomous vehicles need to know their precise location at all times to navigate safely. However, in dense urban areas with tall buildings, trees, and other structures, the signals from GPS satellites can get blocked, making it difficult for the vehicle to determine its position accurately.

To address this problem, the researchers developed a new algorithm that uses images of the sky to detect when the GPS signal is being obstructed. By combining this sky-view information with data from other sensors like inertial measurement units and cameras, the system can continuously track the vehicle's location, even in challenging urban environments.

The system also blends two different positioning techniques - single point positioning (SPP) and real-time kinematic (RTK) - to provide both meter-level accuracy under normal conditions and sub-decimeter precision when more detailed positioning is needed. This makes the system more versatile and reliable for autonomous driving applications.

The researchers have also made the dataset of sky-view images they used publicly available, so other researchers can build upon their work.

Technical Explanation

The paper proposes a sky-view images segmentation algorithm based on Fully Convolutional Network (FCN) for GNSS NLOS detection. Building on this, a novel NLOS detection and mitigation algorithm (S-NDM) is extended to a tightly coupled GNSS, IMU, and visual feature system called Sky-GVIO.

The Sky-GVIO system harmonizes Single Point Positioning (SPP) with Real-Time Kinematic (RTK) methodologies to bolster its operational versatility and resilience. In urban canyon environments, the positioning performance of the S-NDM algorithm is evaluated under different tightly coupled SPP-related and RTK-related models.

The results show that the Sky-GVIO system achieves meter-level accuracy under SPP mode and sub-decimeter precision with RTK, outperforming GNSS/INS/Vision frameworks without the S-NDM component. Additionally, the sky-view image dataset has been made publicly available for further research.

Critical Analysis

The paper presents a comprehensive approach to addressing the challenges of accurate positioning in urban environments. The use of sky-view image segmentation and the S-NDM algorithm to detect and mitigate the effects of NLOS is a novel and promising solution.

However, the paper does not fully address the computational and processing requirements of the proposed system. Integrating multiple sensor modalities and positioning techniques may increase the complexity and resource demands, which could be a limitation for real-world deployment in autonomous vehicles.

Additionally, the paper does not discuss the potential impact of environmental factors, such as weather conditions or seasonal changes, on the performance of the sky-view image segmentation and NLOS detection. Further research may be needed to evaluate the system's robustness under diverse urban environments and varying conditions.

Conclusion

The research presented in this paper offers a significant step towards achieving accurate and continuous positioning for autonomous driving in complex urban environments. The combination of sky-view image analysis, NLOS detection and mitigation, and the harmonization of SPP and RTK techniques demonstrates a promising approach to overcoming the limitations of standalone sensor systems.

By making the sky-view image dataset publicly available, the researchers have opened up opportunities for further exploration and advancements in this field. As autonomous driving technologies continue to evolve, the insights and methods presented in this paper may contribute to the development of more reliable and resilient positioning systems, ultimately enhancing the safety and performance of self-driving vehicles.