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.
4 November 2023 in International Journal of Automotive Technology
In this paper, a method to generate the downward image of current vehicle location using a commercial around view monitor (AVM) is proposed. The proposed system consists of three stages, namely feature tracking, obstacle filtering, and downward view generation. In the feature tracking stage, the Shi-Tomasi corner detection is used and feature tracking is performed with a Kanade-Lucas-Tomasi (KLT) tracker to determine the transformation between AVM images. In the obstacle filtering stage, features on obstacles are filtered by using the difference in a tracking distance between features detected on the ground and features detected on the obstacle. This is performed by using a histogram based on the tracking distance of features. Finally, in the downward view generation stage, transformation between the current AVM image and the previous AVM image is obtained by using the tracked feature pair refined via the aforementioned steps. Specifically, the random sample consensus (RANSAC) method is used to obtain transformations from which the influence of outliers is removed. The downward image of the current vehicle is generated by the obtained transformation and is synthesized to the existing AVM image. The results indicate that the proposed system synthesizes the existing AVM image and the generated downward image of a vehicle in a seamless way by determining the exact pair of matching points between current and the previous AVM images. We believe the proposed system can be utilized in efficient wireless charging system and safe driving system.