Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
98 tokens/sec
GPT-4o
8 tokens/sec
Gemini 2.5 Pro Pro
47 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

PIVO: Probabilistic Inertial-Visual Odometry for Occlusion-Robust Navigation (1708.00894v2)

Published 2 Aug 2017 in cs.CV

Abstract: This paper presents a novel method for visual-inertial odometry. The method is based on an information fusion framework employing low-cost IMU sensors and the monocular camera in a standard smartphone. We formulate a sequential inference scheme, where the IMU drives the dynamical model and the camera frames are used in coupling trailing sequences of augmented poses. The novelty in the model is in taking into account all the cross-terms in the updates, thus propagating the inter-connected uncertainties throughout the model. Stronger coupling between the inertial and visual data sources leads to robustness against occlusion and feature-poor environments. We demonstrate results on data collected with an iPhone and provide comparisons against the Tango device and using the EuRoC data set.

Citations (33)

Summary

  • The paper’s main contribution is a novel probabilistic odometry method that fuses inertial and visual data to tackle occlusions effectively.
  • It employs a discrete-time IMU propagation model and multi-frame visual update, ensuring robust real-time tracking in feature-poor environments.
  • Experimental evaluations show competitive RMSE performance against ORB-SLAM2 and VI-SLAM, enabling reliable navigation on standard smartphone hardware.

PIVO: Probabilistic Inertial-Visual Odometry for Occlusion-Robust Navigation

This paper introduces PIVO, a novel probabilistic inertial-visual odometry system designed for robust navigation in environments prone to occlusion. The method leverages the complementary strengths of visual and inertial sensors to enhance tracking accuracy, particularly in feature-poor and occluded environments. The paper is particularly noteworthy for utilizing standard consumer hardware, namely the low-cost IMU sensors and a monocular camera available on conventional smartphones.

Methodological Framework

PIVO operates within an information fusion framework that integrates data from monocular cameras and MEMS-based inertial measurement units. The model employs a sequential inference approach where the IMU informs the dynamical model, while camera frames update the model with pose estimates in sequences. A key contribution of this work lies in the rigorous treatment of cross-term uncertainties, thereby enhancing the system’s robustness against occlusion and suboptimal visual conditions.

Key features of the PIVO methodology include:

  • IMU Propagation Model: Designed in discrete-time for numerical stability, it incorporates low-cost sensor biases and scale errors. This model allows for closed-form derivatives necessary for the EKF prediction phase, facilitating efficient real-time computation.
  • Pose Augmentation: It maintains a window of past poses using a linear Kalman update. This augmentation of historical poses in the state vector ensures that the information throughput from the sensors is maximized.
  • Visual Update: The visual update stage integrates tracked features over multiple frames, ensuring that all cross-derivative uncertainties are included. This comprehensive approach improves the system's accuracy in environments with sparse feature distribution or temporary obstructions.

Experimental Evaluation

PIVO was tested against existing state-of-the-art methods using the EuRoC MAV dataset. The results demonstrated competitive performance, notably excelling in scenarios with various levels of difficulty. Specifically, it achieved a RMSE in absolute trajectory error that closely matches and occasionally surpasses other prominent techniques such as ORB-SLAM2 and VI-SLAM.

In real-world applications, PIVO was tested on a smartphone, managing to maintain reliable tracking even through challenging conditions like fully occluded camera paths and undefined indoor features. The system functioned effectively both indoors and outdoors, adaptable to pedestrian navigation and wayfinding in complex environments.

Implications and Future Work

The application of PIVO using standard consumer-grade hardware suggests significant practical implications for augmented reality, robotics, and mobile device navigation, providing reliable odometry without the need for specialized equipment. The robust occlusion handling broadens the deployment scenarios for consumer and industrial applications.

Future developments may focus on further refining the model’s computational efficiency, enabling its seamless integration into smaller devices or wearables with limited processing capabilities. Additionally, enhancing multi-sensor fusion techniques to incorporate additional data sources such as GPS or alternative sensor modalities could further bolster PIVO’s applicability across a broader range of scenarios and environments.

In summary, the introduction of PIVO extends the boundaries of what is currently achievable with consumer-grade inertial and visual sensors, offering a promising solution for robust, real-time odometry in diverse and challenging navigation scenarios.

Youtube Logo Streamline Icon: https://streamlinehq.com