GeneA-SLAM2: Dynamic SLAM with AutoEncoder-Preprocessed Genetic Keypoints Resampling and Depth Variance-Guided Dynamic Region Removal

Chinese Conference on Pattern Recognition and Computer Vision (PRCV) 2025

This paper introduces GeneA-SLAM2, an RGB-D SLAM system for dynamic environments. It eliminates dynamic object interference via depth statistical information and enhances keypoint distribution uniformity. Integrated with NGD-SLAM, the system retains real-time performance without a GPU. Experimental results demonstrate that GeneA-SLAM2 maintains high accuracy in dynamic scenes compared to current methods.

News

• 2025-08-23: Accept to PRCV 2025!
• 2025-06-03: Codes released!

Our GeneA-SLAM datasets

We have collected the GeneA-SLAM RGB-D dataset compatible with the TUM format using the ORBBEC Astra sensor, including registered images, trajectories, and point clouds, which can be downloaded from GeneA-SLAM.

Prerequisites

GeneA-SLAM2 is developed based on ORB-SLAM3 and NGD-SLAM. Tested on Ubuntu 18.04; compatible with other platforms. High-performance hardware is recommended for real-time stability. Below are the core dependencies (largely inherited from ORB-SLAM3 and NGD-SLAM with minor optimizations):

PCL (Point Cloud Library)

For 3D point cloud processing, Install via package manager:

sudo apt-get install libpcl-dev

Or build from source: Install Guide

YOLO (Included in Thirdparty folder)

It adopts the C++ version of the YOLO-fastest model from hpc203/yolov34-cpp-opencv-dnn. The model configuration and pre-trained weights are stored in the Thirdparty folder and loaded via OpenCV.

C++11 or C++0x Compiler

Requires a C++11-compatible compiler (for threading/chrono features).

Pangolin

For visualization UI: Install Guide.

OpenCV ≥3.2

For image processing: Install Guide.

Eigen3 ≥3.1.0

For linear algebra (required by g2o): Install Guide.

Thirdparty Libraries (Included)

• DBoW2: Modified for improved place recognition.
• g2o: Modified for optimized non-linear optimization.
  Both are in the Thirdparty folder (BSD licensed).

Python

Required to calculate the alignment of the trajectory with the ground truth. Required Numpy module.

• (win) http://www.python.org/downloads/windows
• (deb) sudo apt install libpython2.7-dev
• (mac) preinstalled with osx

ROS (optional)

We recommend using the automated tool to install the Melodic Desktop Full version of the ROS system on Ubuntu 18.04. The step is optional.

Building GeneA-SLAM2

git clone https://github.com/qingshufan/GeneA-SLAM2

We provide a script build.sh and build_ros.sh to build GeneA-SLAM2.

cd GeneA-SLAM2
chmod +x build.sh
./build.sh
./build_ros.sh

Running datasets

Normal mode

Please modify the dataset path in the following script before running:

cd GeneA-SLAM2
./Examples/RGB-D/rgbd_tum Vocabulary/ORBvoc.txt Examples/RGB-D/TUM3.yaml [path] [path]/associations.txt

ROS mode

We provide a ROS quick-run script. Before running, please modify GeneA_SLAM2_PATH (GeneA-SLAM2 path) and ROSBAG_PATH (ROS bag path) in the script. PLAY_SPEED (playback speed) depends on your computer's performance.

cd GeneA-SLAM2
chmod +x ros_run.sh
./ros_run.sh

LICENSE

ORB-SLAM3 is released under a GPLv3 license. So Our GeneA-SLAM2 are under the GPL-3.0 license.

Acknowledgement

We adapted some codes from some awesome repositories including ORB-SLAM3, NGD-SLAM and ORBSLAM2_with_pointcloud_map. Thanks for making codes publicly available.

Citation

If you find this project useful, please consider citing:

@inproceedings{qing2025geneaslam2, 
      title={GeneA-SLAM2: Dynamic SLAM with AutoEncoder-Preprocessed Genetic Keypoints Resampling and Depth Variance-Guided Dynamic Region Removal}, 
      author={Shufan Qing and Anzhen Li and Qiandi Wang and Yuefeng Niu and Mingchen Feng and Guoliang Hu and Jinqiao Wu and Fengtao Nan and Yingchun Fan},
      booktitle = {Proceedings of the Chinese Pattern Recognition and Computer Vision (PRCV)},
      journal={arXiv preprint arXiv:2506.02736},
      year={2025}
}