Abstract
Many classic visual monocular SLAM (simultaneous localization and mapping) systems have been developed over the past decades, yet most of them fail when dynamic scenarios dominate. DM-SLAM is proposed for handling dynamic objects in environments based on ORB-SLAM2. This article mainly concentrates on two aspects. Firstly, we proposed a distribution and local-based RANSAC (Random Sample Consensus) algorithm (DLRSAC) to extract static features from the dynamic scene based on awareness of the nature difference between motion and static, which is integrated into initialization of DM-SLAM. Secondly, we designed a candidate map points selection mechanism based on neighborhood mutual exclusion to balance the accuracy of tracking camera pose and system robustness in motion scenes. Finally, we conducted experiments in the public dataset and compared DM-SLAM with ORB-SLAM2. The experiments corroborated the superiority of the DM-SLAM.
Highlights
The main task of visual SLAM is to estimate the pose of the camera and reconstruct the three-dimensional structure of the environments in the field of view of continuous frames
In order to show the effectiveness of the DLRSAC algorithm in more detail, we run DM-SLAM on a public data set, and perform DLRSAC experiments
We propose a DM-SLAM system based on ORB-SLAM2 to adapt to dynamic environments
Summary
The main task of visual SLAM (simultaneous localization and mapping) is to estimate the pose of the camera and reconstruct the three-dimensional structure of the environments in the field of view of continuous frames. Visual-based autonomous robots are able to acquire camera poses and environmental information to perform complex tasks such as robot navigation, human–robot interaction, and path planning. Binocular, and RGB-D cameras can be used to implement visual SLAM. Due to their acquisition of depth of environments directly, binocular-based and RGB-D-based SLAM are more robust than monocular SLAM. Despite some remarkable results such as ORB-SLAM2 [2], LSD [3], DSO [4] in visual Monocular cameras have their own inherent limitations (such as inability to observe scale and state initialization), with respect to size, power, and cost [1], monocular cameras have broad application scenarios.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
