Basic introduction to VINS
VINS-Mono and VINS-Mobile are open source monocular visual inertial navigation SLAM solutions from Hong Kong University of Science and Technology Shen Shaojiao. Published in "IEEE Transactions on Robotics" in 2017. In addition, the latest paper of VINS won the iROS 2018 Best Student Paper Award. It is based on optimization and sliding window VIO, using IMU pre-integration to build a tightly coupled frame, and also has automatic initialization, online external parameter calibration, repositioning, closed-loop detection, and global pose map optimization functions.
VINS-Mono is a real-time SLAM framework based on the monocular visual inertial system. It is currently a very advanced monocular VIO algorithm and a classic in the integration of vision and IMU. Its positioning accuracy can be comparable to OKVIS, and it has better positioning accuracy than OKVIS. Complete and robust initialization and closed-loop detection process, the code runs on Linux and is fully integrated with ROS. VINS-Mono is mainly used for state estimation and feedback control of autonomous drones, but it can also provide precise positioning for AR applications. VINS-Mobile can run on the iOS system. It is worth mentioning that Mr. Shaojie Shen has open sourced the ROS and iOS versions. The front-end is very simple and the code is very clear. It is worth learning. Link: https://arxiv.org/pdf/ 1708.03852.pdf .
VINS's overall system framework diagram
The front end is based on the KLT tracking algorithm, the back end is based on sliding window optimization (using the ceres library), and the loop detection based on DBoW.
The overall framework is divided into five parts, as shown in the figure above:
1. Measuremen Preprocessing: Observation data preprocessing, including image data tracking IMU data pre-integration;
2. Initialization: Initialization, including simple visual initialization and visual inertia joint initialization;
3. Local Visual-Inertia BA and Relocalization: local BA joint optimization and relocalization, including a BA optimization model based on a sliding window;
4. Global Pose Graph Optimization: Global Pose Graph Optimization, which only optimizes the global pose;
5. Loop detection: loop detection.
Why does VINS adopt the "vision + IMU" fusion?
A single sensor cannot be applied to all scenes. For example, the visual sensor works well in most scenes with rich textures, but if it encounters scenes with fewer features such as glass and white walls, it basically cannot work. Fusion can achieve the ideal positioning effect.
MYNT’s binocular camera uses a fusion scheme of “binocular + inertial navigation + structured light”
Although the long-term use of the IMU has a very large cumulative error, in a short time, its relative displacement data has a high accuracy, so when the vision sensor fails, fusing the IMU data can improve the accuracy of its positioning. At the same time, the complementary characteristics of vision and inertial measurement make them particularly suitable for fusion, and robustness and accurate positioning and mapping are the main requirements that any mobile robot needs to address. In addition, these two kinds of sensors are present in most smart phones, and the fusion can effectively solve the simultaneous visual-inertial positioning and mapping on mobile phones.
Bu Xiaoyi concluded:
- The integration of vision and IMU can take advantage of the higher sampling frequency of the IMU to increase the output frequency of the system.
- The fusion of vision and IMU can improve the robustness of vision, such as visual SLAM because of the wrong results of certain sports or scenes.
- The integration of vision and IMU can effectively eliminate the integral drift of IMU.
- The fusion of vision and IMU can correct the Bias of IMU.
- The fusion of monocular and IMU can effectively solve the problem of unobservable monocular scale.
MYNT binocular camera standard version runs VINS actual test:
Compared with the OKVIS solution, VINS is faster to build and has a lower CPU usage, making it more suitable for friends to get started quickly.
WINES-Fusion
Recently , the open source VINS-Fusion of Hong Kong University of Science and Technology has won the attention of many friends, and I am very honored to recommend our camera~
VINS-Fusion is an optimized multi-sensor state estimator that can realize precise self-positioning for autonomous applications (drones, cars, and AR/VR). VINS-Fusion is an extension of VINS-Mono and supports multiple types of visual inertial sensors (monocular camera + IMU, binocular camera + IMU, even binocular camera only). The open source project team also showed examples of modules that integrate VINS and GPS.
Features are as follows:
Multi-sensor support (stereo camera / mono camera + IMU / stereo camera + IMU)
Online spatial calibration (conversion between camera and IMU)
Online time calibration (time offset between camera and IMU)
Visual loop closure
My Binocular Camera Standard Edition ran VINS-Fusion actual test:
VINS future direction
The natural complementarity of cameras and IMUs and the popularity of smart phones have made the visual inertial odometer VIO very popular in recent years. Apple's ARKit and Google's ARCore are both typical applications of VIO. VIO provides an effective direction for the future miniaturization and low cost of SLAM, and combined with the sparse direct method, it is expected to achieve good SLAM or VO effects on low-end hardware, which is very promising in the future.
The content of the article is partly from an open class held by XiaoMi Intelligent & Deep Blue Academy:
SMART | How to select VSLAM technology www.shenlanxueyuan.com
Interested friends can go to watch the full content. (The website needs to be registered and logged in to view)
More SLAM learning resources:
My Smart | Visual Odometer (VO) study notes zhuanlan.zhihu.com My Little Smart | ORB-SLAM study notes zhuanlan.zhihu.com My Little Smart | Learn how to get started with SLAM? zhuanlan.zhihu.com small seek intelligence: small seek intelligent | OKVIS study notes zhuanlan.zhihu.com
Everyone is welcome to like it~ The people who like it are the most beautiful~
Friends who want to buy can click here: