Lichen Wang, Naji Khosravan, Sai Sateesh Gudapati
Granted U.S. Invention Patent No. 2025/0166256A1
We build a tool that (1) finds buildings most like a chosen property by comparing floor plans, locations, and other features, and (2) flags the key differences to suggest practical upgrades, ranked by real user-interest data. The results plug straight into design dashboards, maps, or autonomous navigation systems.
[PDF]
Naji Khosravan, Lichen Wang, Sing Bing Kang
Granted U.S. Invention Patent No. 11830135B1
We propose pipelines capable of evaluating similarities across homes by jointly considering various home attributes (e.g., floor plans, locations, sizes, etc.). Multiple machine learning and computer vision algorithms, including visual feature extraction, connectivity analysis, and graph neural networks, are utilized. Additionally, graphical user interfaces are designed to provide explicit and efficient interaction with potential users.
[PDF] [Google Patent]
Eric M. Penner, Naji Khosravan, Sing Bing Kang, Lichen Wang, Zachary S. Bessinger
Granted U.S. Invention Patent No. 2024/0096097A1
We have designed a pipeline capable of automatically generating building information from various attributes. This pipeline also collects building attributes from resources such as floor plans, photos, and panorama images. Furthermore, we propose multiple practical applications, including updating and/or validating existing building descriptions, finding/searching for similar buildings, and displaying on client devices through corresponding graphical user interfaces.
[PDF] [Google Patent]
Bo Zong, Haifeng Chen, Lichen Wang
Granted U.S. Invention Patent No. 20210248425
We propose a Graph-Based Reinforced Text Representation Learning (GRTR) approach that enhances tasks such as natural language processing, text understanding, and decision-making. GRTR utilizes a reinforcement learning module to automatically navigate through the given text, enabling it to identify the most informative clues, such as key words, and provide concise text representations.
[PDF] [Google Patent] [Research Paper]
Bo Zong, Haifeng Chen, Lichen Wang
Granted U.S. Invention Patent No. 20210089652
A stochastic subgraph sampling strategy is proposed, which evaluates the similarities of given graphs quickly and efficiently. Then, methods and systems are designed for detecting abnormal app behaviors. Specifically, syscall graphs are derived from app behaviors, then vector representations are learned via our method. Abnormal actions are determined, and security actions are performed responsively.
[PDF] [Google Patent] [Research Paper]
Lichen Wang, Yan Zhang, Kevin O'Connell
Granted U.S. and International Invention Patent No.11010915
We disclose 3D data processing methods for dynamic container auto-configuration. The systems capture 3D images (point clouds) of the containers, and the methods process the data using a specifically designed pipeline which includes denoising, background removal, recognition, and localization algorithms. The highly accurate results obtained can be utilized for various downstream tasks, such as load estimation.
[PDF-US] [[PDF-CN]] [Google Patent]
Yan Zhang, Kevin O'Connell, Jay Williams, Lichen Wang
Granted U.S. and International Invention Patent No.11010915
We propose a robust point cloud-based container detection system. The system utilizes various algorithms including 3D/2D image conversion, image matching, point cloud segmentation, interior and exterior metrics, and orientation estimation. Through an iterative process, our system accurately determines the location of the container. It exhibits stable and accurate performance even in scenarios with high levels of noise and extreme container poses.
[PDF-US] [PDF-CN] [Google Patent]
Lichen Wang, Min Wu, Qinglin Liu
Granted China Invention Patent No. CN201511025257.3
We propose a method and system for frost detection in an evaporator (e.g., refrigerator). The system consists of multiple cameras that capture images of the evaporator from different views. Image processing algorithms are designed to extract features that accurately and comprehensively reflect the frost condition. This information is used to establish the appropriate start and end time points for the subsequent defrosting process.
[PDF-CN]
Zhenshen Qu, Lichen Wang, Wenhua Jiao, et al.
Granted China Invention Patent No. CN2013102084539
We proposed systems and methods for detecting foreign substances in intravenous bottles, with the aim of replacing the manual inspection process. The system incorporates specifically designed computer vision hardware, such as illumination and mechanical structures for imaging procedures. Advanced image processing algorithms, including denoising, region proposal, and detection, are implemented to ensure accurate and efficient detection of foreign substances.