DYMRO-SLAM: A Robust Stereo Visual SLAM for Dynamic Environments Leveraging Mask R-CNN and Optical Flow

Simultaneous localization and mapping (SLAM) enables robots to localize in uncertain environments and has been widely used in the field of robotics. However, traditional vision SLAM systems usually assume that the observation environment is stationary, which results in unsatisfactory performance in dynamic scenarios. To overcome this challenge, we propose a DYMRO-SLAM system which incorporates a Mask R-CNN instance segmentation network and an optical flow tracking algorithm based on ORB-SLAM3 to enhance the localization performance in dynamic environments. Specifically, the system separates dynamic targets from the stationary background during image processing stage to improve the accuracy of its localization in dynamic scenes; When processing key and non-keyframes, the system adopts feature point tracking and optical flow tracking, respectively, which greatly reduces computation time of descriptors and thus improves the real-time performance. Finally, we validate the performance on datasets from TUM, EuRoC and our own datasets, and the experiment results verifies that our DYMRO-SLAM system has significant performance gains in terms of localization accuracy and runtime efficiency.