Moving object removal for robust visual SLAM in dynamic environment

Visual 3D Simultaneous Localization And Mapping (SLAM) is an important technique to reconstruct a 3D space which helps in the navigation of mobile robots. The classical SLAM systems assume that the environment is static. In dynamic environment, these SLAM systems work in a random manner and affects the SLAM system which degrades the 3D reconstruction as the camera motion estimation gets distorted due to dynamic objects. Due to this reason, in real-time scenario, a mobile robot will have difficulties to navigate in dynamic environments. In this paper, we have proposed a more robust visual SLAM in dynamic environment by eliminating the dynamic objects using Convolutional Neural Network (CNN) and optical flow methods. Our proposed method can generate Sparse 3D maps of dynamic environments by removing the dynamic objects with a robust and more accurate manner.