3 results about "Shape context" patented technology

This invention discloses a method and system for point cloud completion of complex plant structures based on an end-to-end network. The method constructs a hierarchical geometric encoder to extract and fuse multi-scale geometric features and global shape context features from an incomplete input tree point cloud. A seed generator then generates a coarse point cloud containing the main skeleton of the tree. A cascaded tree growth module progressively upsamples the coarse point cloud, and dynamic deformation constraints ensure that the upsampling process conforms to the natural growth pattern of trees, ultimately outputting a structurally complete and realistically shaped tree point cloud. Furthermore, this invention addresses the scarcity of real, complete tree point cloud data by fusing UAV and ground-based LiDAR scanning data, providing high-quality supervisory data for model training. This effectively solves the problems of insufficient semantic understanding, loss of detail, and morphological distortion in existing point cloud completion methods when dealing with complex tree branch structures, significantly improving completion accuracy and morphological fidelity.

The invention belongs to the technical field of computer vision, and particularly relates to an end-to-end shape recognition method and system based on a complex network. The method comprises the following steps: carrying out dense discretization on the edge of a to-be-identified image to obtain an edge description point set; selecting a series of key points on the obtained edge description point set at equal intervals; extracting local features of a key point surrounding edge description point set by using a shape context; taking each key point as a node, defining edges in the network according to spatial correlation among the key points, and constructing a multi-layer complex network by adopting a dynamic evolution strategy; local features and topological features of the complex network are processed through a graph convolutional network, and robust expression of the shape classification task is obtained; and performing classification through a full connection layer network to obtain a classification result, and realizing end-to-end shape recognition. According to the method, the adaptability and robustness of the method to different shape classification tasks are enhanced.

This invention provides a method for road marking extraction and classification based on a local shape Transformer network. The method includes: acquiring an original LiDAR point cloud and performing non-ground point filtering preprocessing to obtain a road surface point cloud; inputting the roadside point cloud into a Transformer-based encoder network, and obtaining semantic features containing local geometric details and global shape context through multi-stage downsampling and feature extraction; using a decoder network to upsample and recover the semantic features, fusing multi-scale information with skip connections, and outputting the road marking category prediction result for each point through a classification head. This invention improves the extraction accuracy and boundary integrity of fine markings such as lane lines and arrows.