Search Results (4)

Lidar point clouds are crucial for autonomous driving, but their sparsity and scale variations pose challenges for instance segmentation. In this paper, we propose LCPSNet, a Light Detection and Ranging (LiDAR) channel-aware point segmentation network designed to handle distance-dependent sparsity and scale variation in point clouds. A top-down FPN is adopted, where high-level features are progressively upsampled and fused with shallow layers. The fused features at 1/16, 1/8, and 1/4 are further aligned to a common BEV/polar grid and processed by the Local Perception Module (LPM), which applies cross-scale, position-dependent weighting to enhance intra-object coherence and suppress interference. The Inter-Channel Correlation Module (ICCM) employs ball queries to model spatial and channel correlations, computing an inter-channel similarity matrix to reduce redundancy and highlight valid features. Experiments on SemanticKITTI and Waymo show that LPM and ICCM effectively improve local feature refinement and global semantic consistency. LCPSNet achieves 70.9 PQ and 77.1 mIoU on SemanticKITTI, surpassing mainstream methods and reaching state-of-the-art performance. Full article

Point cloud registration is pivotal across various applications, yet traditional methods rely on unordered point clouds, leading to significant challenges in terms of computational complexity and feature richness. These methods often use k-nearest neighbors (KNN) or neighborhood ball queries to access local neighborhood information, which is not only computationally intensive but also confines the analysis within the object’s boundary, making it difficult to determine if points are precisely on the boundary using local features alone. This indicates a lack of sufficient local feature richness. In this paper, we propose a novel registration strategy utilizing ordered point clouds, which are now obtainable through advanced depth cameras, 3D sensors, and structured light-based 3D reconstruction. Our approach eliminates the need for computationally expensive KNN queries by leveraging the inherent ordering of points, significantly reducing processing time; extracts local features by utilizing 2D coordinates, providing richer features compared to traditional methods, which are constrained by object boundaries; compares feature similarity between two point clouds without keypoint extraction, enhancing efficiency and accuracy; and integrates image feature-matching techniques, leveraging the coordinate correspondence between 2D images and 3D-ordered point clouds. Experiments on both synthetic and real-world datasets, including indoor and industrial environments, demonstrate that our algorithm achieves an optimal balance between registration accuracy and efficiency, with registration times consistently under one second. Full article

Recent advances in sensor technologies, in particular video-based human detection, object tracking and pose estimation, have opened new possibilities for the automatic or semi-automatic per-frame annotation of sport videos. In the case of racket sports such as tennis and padel, state-of-the-art deep learning methods allow the robust detection and tracking of the players from a single video, which can be combined with ball tracking and shot recognition techniques to obtain a precise description of the play state at every frame. These data, which might include the court-space position of the players, their speeds, accelerations, shots and ball trajectories, can be exported in tabular format for further analysis. Unfortunately, the limitations of traditional table-based methods for analyzing such sport data are twofold. On the one hand, these methods cannot represent complex spatio-temporal queries in a compact, readable way, usable by sport analysts. On the other hand, traditional data visualization tools often fail to convey all the information available in the video (such as the precise body motion before, during and after the execution of a shot) and resulting plots only show a small portion of the available data. In this paper we address these two limitations by focusing on the analysis of video-based tracking data of padel matches. In particular, we propose a domain-specific query language to facilitate coaches and sport analysts to write queries in a very compact form. Additionally, we enrich the data visualization plots by linking each data item to a specific segment of the video so that analysts have full access to all the details related to the query. We demonstrate the flexibility of our system by collecting and converting into readable queries multiple tips and hypotheses on padel strategies extracted from the literature. Full article

Semantic segmentation for 3D point clouds plays a critical role in the construction of 3D models. Due to the sparse and disordered natures of the point clouds, semantic segmentation of such unstructured data yields technical challenges. A recently proposed deep neural network, PointNet, delivers attractive semantic segmentation performance, but it only exploits the global features of point clouds without incorporating any local features, limiting its ability to recognize fine-grained patterns. For that, this paper proposes a deeper hierarchical structure called the high precision range search (HPRS) network, which can learn local features with increasing contextual scales. We develop an adaptive ball query algorithm that designs a comprehensive set of grouping strategies. It can gather detailed local feature points in comparison to the common ball query algorithm, especially when there are not enough feature points within the ball range. Furthermore, compared to the sole use of either the max pooling or the mean pooling, our network combining the two can aggregate point features of the local regions from hierarchy structure while resolving the disorder of points and minimizing the information loss of features. The network achieves superior performance on the S3DIS dataset, with a mIoU declined by 0.26% compared to the state-of-the-art DPFA network. Full article