Search

Your search keyword '"Wu, Hangbin"' showing total 197 results
Start Over Author "Wu, Hangbin"
197 results on '"Wu, Hangbin"'

32. Recognition of Indoor Scenes Using 3-D Scene Graphs

Author

Yue, Han, Lehtola, Ville, Wu, Hangbin, Vosselman, George, Li, Jincheng, and Liu, Chun

Abstract

Scene recognition is a fundamental task in 3-D scene understanding. It answers the question, “What is this place?” In an indoor environment, the answer can be an office, kitchen, lobby, and so on. As the number of point clouds increases, using embedded point information in scene recognition becomes computationally heavy to process. To achieve computational efficiency and accurate classification, our idea is to use an indoor scene graph that represents the 3-D spatial structures via object instances. The proposed method comprises two parts, namely: 1) construction of indoor scene graphs leveraging object instances and their spatial relationships and 2) classification of these graphs using a deep learning network. Specifically, each indoor scene is represented by a graph, where each node represents either a structural element (like a ceiling, a wall, or a floor) or a piece of furniture (like a chair or a table), and each edge encodes the spatial relationship between these elements. Then, these graphs are used as input for our proposed graph classification network to learn different scene representations. The public indoor dataset, ScanNet v2, with 625.53 million points, is selected to test our method. Experiments yield good results with up to 88.00% accuracy and 82.30% F1 score in the fixed validation dataset and 90.46% accuracy and 81.45% F1 score in the ten-fold cross-validation method; moreover, if some indoor objects cannot be successfully identified, the scene classification accuracy depends sublinearly on the rate of missing objects in the scene.

Published
2024
Full Text
View/download PDF

40. A real-time map matching method for road network using driving scenario classification

Author

FU Chen, HUANG Shengke, TANG Yan, WU Hangbin, LIU Chun, YAO Lianbi, and HUANG Wei

Subjects
gnss trajectory, Mathematical geography. Cartography, driving scenario classification, GA1-1776, map-matching
Abstract

Real-time map matching plays a critical role in intelligent transportation and autonomous driving. For complex road networks like elevated roads and overpasses, existing real-time matching algorithms have relatively lower accuracy due to the interference of parallel roads. Thus, a real-time map matching method combined with driving image classification is proposed. When the vehicle nears the elevated roads, the current trajectory point is matched by combining the scenario classification result with the vehicle's heading direction, the distance to the road segment, and the adjacency with the previous matching segment. For the experiment, three trajectories with high GNSS sampling rates were collected in Shanghai. Three indicators (match rate, recall, and precision) are used to evaluate the matching performance. The results show that the average matching rate, recall, and precision of the proposed method are 96.86%, 97.17%, 93.46%, which outperform the traditional real-time matching methods. As the sampling interval increases, the proposed method still performs well with three indicators. Comparing the matching results in complex areas such as elevated roads and intersections, as well as comparing the matching time, latency and memory consumption, this method can maintain good matching results.

Published
2021

41. Visual odometry optimizing bounded with semantic elements association in dynamic scenes

Author

SHAO Xiaohang, WU Hangbin, LIU Chun, CHEN Chen, CAI Tianchi, and CHENG Fanjin

Subjects
visual odometry, dynamic scenes, slam, Mathematical geography. Cartography, semantic association, GA1-1776, semantic segmentation
Abstract

Cameras can help to make low-cost positioning and environment perceiving in autopilot but dynamic objects give negative effects to visual odometry. This paper gives a model to optimize it based on semantic elements association. It uses such techniques as objects detection and semantic segmentation to identify semantic objects and distinguish dynamic semantic elements(DSE) from static ones. Then it filters bad keypoints by a dynamic feature mask in visual positioning. In practice, this proposed method detects DSE even when there are objects with duality of moving and static. In an experiment on campus roads, negative influences were found especially when a robot turning around or a moving object crossing its camera view. It showed that average accuracy in detecting DSE was 87% and the largest difference between trajectories before and after semantic association optimizing in sub-sequences was 2.463 m. Compared to ground truth, the RMSE of proposed method dropped by 38%.

Published
2021

Catalog

Books, media, physical & digital resources
See catalog results