UMS-VINS: United Monocular-Stereo Features for Visual-Inertial Tightly Coupled Odometry (2303.08550v1)

Abstract: This paper introduces the united monocular-stereo features into a visual-inertial tightly coupled odometry (UMS-VINS) for robust pose estimation. UMS-VINS requires two cameras and a low-cost inertial measurement unit (IMU). The UMS-VINS is an evolution of VINS-FUSION, which modifies the VINS-FUSION from the following three perspectives. 1) UMS-VINS extracts and tracks features from the sub-pixel plane to achieve better positions of the features. 2) UMS-VINS introduces additional 2-dimensional features from the left and/or right cameras. 3) If the visual initialization fails, the IMU propagation is directly used for pose estimation, and if the visual-IMU alignment fails, UMS-VINS estimates the pose via the visual odometry. The performances on both public datasets and new real-world experiments indicate that the proposed UMS-VINS outperforms the VINS-FUSION from the perspective of localization accuracy, localization robustness, and environmental adaptability.

• The paper presents a novel odometry algorithm that fuses monocular-stereo visual features with IMU data to enhance pose estimation accuracy.
• Advanced sub-pixel feature extraction and tracking, along with additional 2D features, significantly improve localization precision.
• Adaptive switching between visual-IMU alignment and direct IMU propagation ensures reliable performance across diverse and challenging environments.

"UMS-VINS: United Monocular-Stereo Features for Visual-Inertial Tightly Coupled Odometry" introduces a novel odometry system leveraging the united monocular-stereo feature sets combined with inertial measurements for robust and accurate pose estimation. This paper is an extension and evolution of the VINS-FUSION system and proposes significant improvements in its methodology to enhance performance in various environments.

Key Innovations and Methodological Enhancements:

1. Sub-Pixel Feature Extraction and Tracking:
   • UMS-VINS adopts advanced techniques to extract and track features on a sub-pixel level, which significantly enhances the precision of feature positions. This fine-grained approach allows for a more accurate representation of environmental features, leading to more reliable pose estimations.
2. Incorporation of Additional 2D Features:
   • The system integrates extra 2-dimensional features sourced from either the left or right cameras. This inclusion is critical in ensuring that the odometry process utilizes a richer set of visual data, improving feature robustness and spatial understanding.
3. Enhanced Visual-IMU Integration:
   • One of the central features of UMS-VINS is its adaptive approach when visual initialization or visual-IMU alignment faces difficulties. If visual initialization fails, the system employs direct IMU propagation to estimate the pose. Conversely, if visual-IMU alignment is unsuccessful, UMS-VINS switches to estimate the pose using visual odometry alone. This flexibility enhances the system’s robustness in diverse and challenging environments.

Performance and Evaluation:

• Localization Accuracy:

The UMS-VINS system has demonstrated superior localization accuracy compared to its predecessor, VINS-FUSION. This improvement stems mainly from the enhanced sub-pixel feature tracking and enriched 2D feature set, which together provide a more precise and detailed environmental mapping.

• Localization Robustness:

Through rigorous testing on both public datasets and new real-world experiments, UMS-VINS has shown marked improvements in maintaining reliable localization, even under conditions that typically challenge visual-inertial odometry systems. The ability to adaptively switch between different pose estimation strategies plays a crucial role in this robustness.

• Environmental Adaptability:

UMS-VINS has been tested across various environments, evidencing its adaptability and effectiveness across different visual and inertial scenarios. The system's design allows it to adjust dynamically to different types of input data and environmental conditions, maintaining high performance in both structured and unstructured settings.

In conclusion, UMS-VINS represents a significant advancement in the field of visual-inertial odometry by integrating united monocular-stereo features with adaptive IMU usage. The outcome is a system that offers increased accuracy, robustness, and versatility, marking a step forward from previous odometry solutions such as VINS-FUSION.