An adaptive visual Dynamic-SLAM method based on fusing the semantic information

Jichao Jiao,Chenxu Wang,Ning Li,Zhongliang Deng,Wei Xu

DOI: https://doi.org/10.1109/jsen.2021.3051691

IF: 4.3

2021-01-01

IEEE Sensors Journal

Abstract:The SLAM problem in dynamic scenes is regarded as a challenge. This article proposes a novel SLAM framework for dynamic environments, which combines neural network and motion information of dynamic objects, making the system more adaptable to dynamic scenes. Specifically, we adopt a fast object detection network, tightly couple the results of the object detection with the geometric information in the SLAM system. Then the feature point extracted from the image is associated with a dynamic probability. By utilizing the feature points which tend to be static, the localization result can be greatly improved in dynamic environments. We perform the experiments both on the public data set and the real environment. The result can demonstrate that the proposed method greatly improves the localization accuracy in a dynamic environment. An open-source version of the source code is available.

engineering, electrical & electronic,instruments & instrumentation,physics, applied

What problem does this paper attempt to address?

This paper attempts to address the problem of SLAM (Simultaneous Localization and Mapping) in dynamic scenes. Traditional visual SLAM methods usually assume that the environment is static, but in real-world environments, there are often moving objects that can affect the accuracy of localization and mapping. Therefore, this paper proposes a new SLAM framework that combines neural networks and motion information of dynamic objects to make the system more adaptable in dynamic environments.

Specifically, this method employs a fast object detection network, tightly integrating the object detection results with the geometric information of the SLAM system, and assigns a dynamic probability to the feature points in the image. By utilizing those feature points that tend to be stationary, the localization accuracy in dynamic environments can be significantly improved. Experimental results show that this method can significantly improve localization accuracy in both public datasets and real-world environments.