Improved Real-Time Monocular SLAM Using Semantic Segmentation on Selective Frames

Monocular simultaneous localization and mapping (SLAM) is emerging in advanced driver assistance systems and autonomous driving, because a single camera is cheap and easy to install. Conventional monocular SLAM has two major challenges leading inaccurate localization and mapping. First, it is challenging to estimate scales in localization and mapping. Second, conventional monocular SLAM uses inappropriate mapping factors such as dynamic objects and low-parallax areas in mapping. This paper proposes an improved real-time monocular SLAM that resolves the aforementioned challenges by efficiently using deep learning-based semantic segmentation. To achieve the real-time execution of the proposed method, we apply semantic segmentation only to downsampled keyframes in parallel with mapping processes. In addition, the proposed method corrects scales of camera poses and three-dimensional (3D) points, using estimated ground plane from road-labeled 3D points and the real camera height. The proposed method also removes inappropriate corner features labeled as moving objects and low parallax areas. Experiments with eight video sequences demonstrate that the proposed monocular SLAM system achieves significantly improved and comparable trajectory tracking accuracy, compared to existing state-of-the-art monocular and stereo SLAM systems, respectively. The proposed system can achieve real-time tracking on a standard CPU potentially with a standard GPU support, whereas existing segmentation-aided monocular SLAM does not.

I. Introduction

Simultaneous localization and mapping (SLAM) techniques have been evolving and widely applied to advanced driver assistance systems (ADAS) and autonomous driving systems. While SLAM approaches using light detection and ranging (LiDAR) sensors are accurate, the cost of LiDAR sensors is high and they have not been widely used in commercial products. Visual SLAM systems that use camera(s) are a popular alternative to LiDAR-based SLAM. Monocular SLAM systems that use a single camera are attractive as they are cheap and easy to install. Monocular SLAM was initially suggested with filter-based approaches [1], [2], [3], [4]. The filter-based methods are computationally inefficient, since both localization and mapping run on every frame [5]. To resolve the issue of filter-based methods, keyframe-based approaches [6], [7], [8] (see other references in [5]) run the mapping process only on selective frames, called keyframes, while the localization process estimates a camera pose in every frame. The keyframe-based SLAM improved the localization accuracy and computational efficiency of filter-based methods [9], and became the de facto standard in monocular SLAM [5].