refer to: https://docs.openvins.com/

Executive Summary

This report outlines the systematic implementation of a real-time Visual Inertial Odometry (VIO) system utilizing the ZED2i camera. The system is designed to provide high-accuracy motion tracking for robotics and augmented reality applications. Key phases of the implementation included setting up a compatible hardware and software environment, installing and configuring necessary software packages, and calibrating the system to ensure optimal performance.

1. System Requirements

The implementation was carried out on a system equipped with Ubuntu 20.04 and ROS Noetic, chosen for their stability and compatibility with the ZED2i camera. The hardware specifications were selected to support the intensive computational demands of real-time odometry and data processing.

2. Software Installation and Setup

The ZED Software Development Kit (SDK) was installed to facilitate camera integration. Following this, the ZED ROS package was deployed to enable the camera’s data to be utilized within the ROS ecosystem. These installations ensured that the camera’s video and depth data could be effectively captured and published as ROS topics.

3. Calibration Process

Calibration was performed using Kalibr, alongside the Allan variance tool integrated within ROS, to calibrate the ZED2i camera and its inertial measurement units (IMU). This calibration was crucial for refining the accuracy of the sensors and ensuring that the VIO system could reliably interpret sensory data in real time.

4. Visual-Inertial System Integration

OpenVINS, an open-source visual-inertial estimator, was installed to process the visual and inertial data. The software was configured specifically for our hardware and the output from the calibration phase. This integration facilitated the real-time estimation of position, orientation, and velocity, which are pivotal for the precise navigation and interaction capabilities required in robotics and augmented reality.

5. Operational Results

Upon running the integrated system, initial tests demonstrated successful real-time tracking with minimal latency. The system’s ability to accurately track motion in a variety of conditions was verified through several experimental setups, highlighting the efficacy of the calibration and the robustness of the VIO algorithm.