![]() ![]() ![]() However, for indoor applications, existing learning-based methods cannot directly infer semantic labels of point-cloud frames because there is a lack of frame-level labeled datasets for training. As a multi-task strategy, semantic-oriented mapping employs semantic information to filter dynamic objects or track dynamic keypoints to further improve the accuracy and robustness of the mapping task in an outdoor environment. Integrating multiple tasks into a framework has been proven to be effective in many applications for instance, multi-task learning simultaneously uses different types of data to learn multimodal mapping relationships or integrated features. However, the mapping process using this strategy limits the accuracy of the two latter tasks. Semantic labeling and 3D modeling are performed separately based on offline mapped point clouds. In most previous studies, mapping, semantic labeling, and 3D modeling have been performed separately. Numerous studies have been conducted on the semantic labeling of point clouds using deep neural networks. In recent years, owing to the development of techniques such as virtual reality, artificial intelligence, and autonomous robots, there has been an upward trend in demand for point clouds and models with semantic information. Some applications, such as level-of-detail generation or construction structure rebuilding, do not require intricate models. The level of detail for each model can be selected based on the demands of later-stage applications. Thus, to address this problem, storage-saving models, such as voxels, meshes, planes, and lines, are reconstructed from point clouds to represent scenes. Storing and maintaining such massive data for next-stage applications are a waste of resources. Generally, applications do not require extremely elaborate data. Ī point cloud map generated by applying simultaneous localization and mapping (SLAM) technology using high-precision LiDAR may contain hundreds of millions of points, or even more, in a large scene. LiDAR techniques have been widely adopted for indoor mapping of 3D buildings owing to their high accuracy and good stability. Current commonly used data collection techniques include 3D light detection and ranging (LiDAR), photogrammetry, RGB-D cameras, and stereo cameras. With the rapid development of 3D-sensor technologies, point cloud data of large buildings can be collected quickly and precisely. However, the architectural sketches of existing buildings are often outdated or missing, especially for frequently remodeled or rebuilt buildings. Buildings are essential elements of smart cities, and producing their sketches is of considerable importance. Owing to the advent of the smart city era, there is a growing demand to acquire and update the spatial information of large buildings. ![]() The experimental results show that the proposed method achieves better results than the state-of-the-art methods that separately perform one of the two tasks. Finally, the optimized poses are used to integrate semantic frames and line structures to generate a point cloud map and 3D line model of buildings. In the subsequent pose optimization process, the initial poses are optimized under the constraints of the structural planes. Then, these frames are used to estimate the initial poses of a 3D sensor for data collection. Subsequently, point cloud frames with semantic labels are used to extract the structural planes of buildings, followed by the generation of line structures from the planes. First, our framework uses a deep-learning-assisted method to perform frame-level point cloud semantic labeling. To address this issue, we propose a framework to cooperatively perform the three tasks of semantic labeling, mapping, and 3D modeling of point clouds. ![]() Studies are currently being conducted on semantic labeling and modeling based on offline mapped point clouds, in which, the performance is strongly limited by the mapping process. Point clouds and models with semantic information facilitate various indoor automation, ranging from indoor robotics to emergency responses. ![]()
0 Comments
Leave a Reply. |