GOReloc

This repository contains the implementation of our RAL paper: GOReloc: Graph-based Object-Level Data Association for Relocalization(IEEE, Arxiv). The article introduces a novel method for object-level relocalization of robotic systems. It determines the pose of a camera sensor by robustly associating the object detections in the current frame with 3D objects in a lightweight object-level map. Object graphs, considering semantic uncertainties, are constructed for both the incoming camera frame and the pre-built map. Objects are represented as graph nodes, and each node employs unique semantic descriptors based on our devised graph kernels. We extract a subgraph from the target map graph by identifying potential object associations for each object detection, then refine these associations and pose estimations using a RANSAC-inspired strategy.

The system overview is as follows:

Installation

The dependencies and installation procedures are consistent with those of our previous repository, VOOM.

Need Install

• Pangolin We use Pangolin for visualization and user interface. Download and install instructions can be found at: https://github.com/stevenlovegrove/Pangolin.
• OpenCV We use OpenCV to manipulate images and features. Download and install instructions can be found at: http://opencv.org. Required at least 2.4.3. Tested with OpenCV 2.4.11 and OpenCV 3.2.
• Eigen3 Required by g2o (see below). Download and install instructions can be found at: http://eigen.tuxfamily.org. Required at least 3.1.0.
• Dlib We use Dlib in the comparison object data association method used by OA-SLAM. Download and install instructions can be found at: https://github.com/davisking/dlib
• Protocol Buffers It is used for Osmap (see below). Download and install instructions can be found at: https://github.com/protocolbuffers/protobuf

Included in the Thirdparty folder

• DBoW2 and g2o We use modified versions of the DBoW2 library to perform place recognition and g2o library to perform non-linear optimizations. Both modified libraries (which are BSD) are included in the Thirdparty folder.
• Json for I/O json files.
• Osmap for map saving/loading. Modified version to handle objects.

Compilation

1. Clone the repository recursively:

   git clone https://github.com/yutongwangBIT/GOReloc.git

2. Build:

   sh build.sh

Data

1. TUM RGBD
2. LM Data Diamond sequences

We use the same detections in JSON files as in VOOM: https://github.com/yutongwangBIT/VOOM/tree/main/Data. If you would like to process your own dataset, please find the Python scripts at: https://github.com/yutongwangBIT/VOOM/tree/main/PythonScripts

Run our system

All command lines can be found at https://github.com/yutongwangBIT/GOReloc/blob/main/script

An example usage on TUM Fr2_desk sequence: (We have uploaded some generated maps, so you can go directly to step 3.)

1. Run VOOM to generate maps with keyframes, map points, and objects:

cd bin/
./rgbd_tum_with_ellipse ../Vocabulary/ORBvoc.txt ../Cameras/TUM2.yaml [Path Source Data fr2_desk] ../Data/fr2_desk/fr2_associated_ours.txt ../Data/fr2_desk/detections_yolov8x_seg_tum_rgbd_fr2_desk_with_ellipse.json null points fr2_desk_all

Here, fr2_desk_all is a folder name to save the map. You can find more JSON files in VOOM at https://github.com/yutongwangBIT/VOOM/tree/main/Data.

2. Move the map folder from bin to Saved_Maps.(there are already some example maps)

3. Run force Relocalization:

./loc_tum ../Vocabulary/ORBvoc.txt ../Cameras/TUM2.yaml [Path Source fr2_person] ../Data/fr2_person/fr2_person_associated.txt ../Data/fr2_person/detections_yolov8x_seg_fr2_person_with_ellipse.json null ../Saved_Maps/fr2_desk_all/map_fr2_desk_all.yaml goreloc tum_reloc 0

Code Explanation

The key components related to this paper include the following:

1. Graph Construction: The Graph class is implemented in Graph.cc. Implementation of how the graph is constructed can be found in Tracking::GrabImageRGBD, where it shows how the graph is built during the tracking phase of RGB-D SLAM. Please note that while we tested the function in RGB-D mode, we actually only use RGB images for relocalization. Consequently, the function can also be compatible with Mono and Stereo modes.
2. Map Saving and Loading: We utilized a third-party library, OSMAP, and extended it to include support for encoding and decoding objects and graphs.
3. GOReloc Function: The core implementation of the GOReloc function can be found in Tracking::GOReloc, corresponding to the methods described in the paper.

Publication

Please cite the corresponding RA-L paper:

    @ARTICLE{wang2024ral,
            author={Wang, Yutong and Jiang, Chaoyang and Chen, Xieyuanli},
            journal={IEEE Robotics and Automation Letters}, 
            title={GOReloc: Graph-based Object-Level Relocalization for Visual SLAM}, 
            year={2024},
            volume={9},
            number={10},
            pages={8234-8241},
            doi={10.1109/LRA.2024.3442560}
    }