A5101933531- Optical measurement and interference techniques
- Advanced Vision and Imaging
- Advanced Neural Network Applications
- Face recognition and analysis
- Advanced Optical Sensing Technologies
- Video Surveillance and Tracking Methods
- Traffic Prediction and Management Techniques
- Transportation Planning and Optimization
- Image Enhancement Techniques
- Digital Media Forensic Detection
- Biometric Identification and Security
- Advanced Image and Video Retrieval Techniques
- Human Mobility and Location-Based Analysis
- ECG Monitoring and Analysis
- Service-Oriented Architecture and Web Services
- Surface Roughness and Optical Measurements
- EEG and Brain-Computer Interfaces
- Cardiac electrophysiology and arrhythmias
- Orbital Angular Momentum in Optics
- Optical Systems and Laser Technology
- Remote Sensing and Land Use
- Robotics and Sensor-Based Localization
- 3D Surveying and Cultural Heritage
- Energy Efficient Wireless Sensor Networks
- Reinforcement Learning in Robotics
Peking University First Hospital
2025
Peking University
2025
Shenzhen Institutes of Advanced Technology
2018-2024
Chinese PLA General Hospital
2024
Peng Cheng Laboratory
2024
Shenzhen Luohu People's Hospital
2024
Shenzhen University
2024
Chinese Academy of Sciences
2013-2023
China Mobile (China)
2019-2023
University of Chinese Academy of Sciences
2023
Face Anti-spoofing gains increased attentions recently in both academic and industrial fields. With the emergence of various CNN based solutions, multi-modal(RGB, depth IR) methods showed better performance than single modal classifiers. However, there is a need for improving reducing complexity. Therefore, an extreme light network architecture(FeatherNet A/B) proposed with streaming module which fixes weakness Global Average Pooling uses less parameters. Our FeatherNet trained by image...
Multispectral pedestrian detection, which consists of a color stream and thermal stream, is essential under conditions insufficient illumination because the fusion two streams can provide complementary information for detecting pedestrians based on deep convolutional neural networks (CNNs). In this paper, we introduced adapted simple efficient one-stage YOLOv4 to replace current state-of-the-art two-stage fast-RCNN multispectral detection directly predict bounding boxes with confidence...
Unlike ordinary computer vision tasks that focus more on the semantic content of images, image manipulation detection task pays attention to subtle information manipulation. In this paper, noise extracted by improved constrained convolution is used as input model instead original obtain traces Meanwhile, dual branch network, consisting a high-resolution and context branch, capture artifacts. general, most leaves artifacts region boundary. A specially designed edge module constructed based...
Cellular traffic prediction is an indispensable part for intelligent telecommunication networks. Nevertheless, due to the frequent user mobility and complex network scheduling mechanisms, cellular often inherits complicated spatial-temporal patterns, making incredibly challenging. Although recent advanced algorithms such as graph-based approaches have been proposed, they frequently model spatial dependencies based on static or dynamic graphs neglect coexisting multiple correlations induced...
Compared with existing depth cameras, such as RGB-D, RealSense and Kinect, stripe-based structured light (SL) has the potential for micrometer-level 3D measurement; this can be attributed to its higher coding capacity. While surface texture, high-reflective region, occlusion remain some of main sources leading degraded reconstruction quality in complex objects, methods that are only based on SL cannot completely solve problems object reconstruction. In paper, we developed an advanced fusion...
3D real-time acquisition plays a vital role in computer graphics and vision. In this paper, we present dynamic IR structured light sensing system with high resolution accuracy for scanning. We adopt the Gray code combined stripe shifting as our acquisition's coding strategy parallelize algorithm via GPU scanning system. Our built-up can capture dense high-precision model sequences speed of 29 Hz. Furthermore, propose practical calibration method to obtain accurate parameters Finally, various...
Mobile network traffic forecasting is one of the key functions in daily operation. A commercial mobile large, heterogeneous, complex and dynamic. These intrinsic features make far from being solved even with recent advanced algorithms such as graph convolutional network-based prediction approaches various attention mechanisms, which have been proved successful vehicle forecasting. In this paper, we cast problem a spatial-temporal sequence task. We propose novel deep learning architecture,...
Speckle dots have the advantage of easy projection, which makes them good candidate features structured light (SL) cameras, such as Kinect v1. However, they generally yield poor accuracy due to block matching. To improve their accuracy, this paper proposes a dot-coded SL, coding information is added into dot distribution. Some are arranged regularly provide easy-to-locate corner features, while others specially designed form different shapes unique identification. A Gaussian-cross module and...
An automatic region proposal network (ARPN) is proposed to generate bounding boxes with confidence scores for far-infrared (FIR) pedestrian detection. The model consists of two parts: first, the are predicted by L2 loss function and a module designed based on convolutional neural network. This simple only has layers, each 1Ã-1 kernel. Second, score map obtained through FIR segmentation feature pyramid taken as levels boxes. To obtain scores, new labelling method also introduced in this paper...
Arrhythmia is a cardiovascular disease that manifests irregular heartbeats. In arrhythmia detection, the electrocardiogram (ECG) signal an important diagnostic technique. However, manually evaluating ECG signals complicated and time-consuming task. With application of convolutional neural networks (CNNs), evaluation process has been accelerated performance improved. It noteworthy CNNs heavily depends on their architecture design, which complex grounded expert experience trial-and-error. this...
Abstract Segmentation of pests is a critical step in using machine vision for field automation tasks. A new method called GaborBoostSVM proposed this paper. The comprises Gabor wavelets‐based feature extraction, AdaBoost‐based selection and SVM‐based pattern recognition algorithm. It performed unsupervised classification images into pest background categories real‐time selective insecticide application. results showed that the capable performing texture‐based consistently high, effectively...
Internet has become the essential platform for access, processing and dissemination of scientific data literature. To greatly enhance academic search, a novel metasearch engine called e-Scholar is presented in this paper, which combined with entity extraction techniques to link relevant databases We demonstrate scalable architecture implementing functions little overhead. also present our innovational approach linking techniques. features significant system performance improvements by...
The rapid development of deep reinforcement learning makes it widely used in multi-agent environments to solve the cooperation problem. However, due instability environments, performance is insufficient when using algorithms train each agent independently. In this work, we use framework centralized training with decentralized execution extend maximum entropy algorithm Soft Actor-Critic (SAC) and proposes MASAC based on framework. Proposed model treats all agents as part environment, can...
To characterize the clinical and imaging features of circumscribed choroidal hemangioma (CCH), to evaluate individualized treatment efficiency photodynamic therapy (PDT), transpupillary thermotherapy (TTT), or their combination, followed by retrobulbar injection betamethasone on CCH resolvement. Forty-nine patients with CCHs who underwent PDT, TTT PDT+TTT treatments were retrospectively analyzed. Their efficacy was compared analyzing change best corrected visual acuity (BCVA), subretinal...
Federated Learning learns a shared model for prediction from the rich and privacy-sensitive training data distributed on clients which are resource-constrained edge devices or isolated cloud clusters with high-security protection. The collected by separated faces non-IID (not Independent Identically Distributed) challenge, requires more communication rounds to achieve an acceptable precision hence costs considerable computing network resources. bias of dataset lot Existing works propose...
This paper presents a partition-based surface registration for 3D morphable model(3DMM). In the 3DMM, it often requires to warp handcrafted template model into different captured models. The proposed method first utilizes landmarks partition then scale each part and finally smooth boundaries. is especially effective when disparity between target huge. experiment result shows perform well than traditional robust local minima.
Three-dimensional measurement of transparent objects has always been a challenging task in the field optics. This is because measurements can be affected by complex optical effects such as transmission, refraction, and multiple reflections. study aims to investigate methods for based on polarization analysis phase deflection techniques improve accuracy reliability. In this article, we introduced principle technique firstly. Then, set up an experimental system using polarized light sources...
In this paper, we propose a real-time, high-precision virtual meeting system based on infrared structured light 3D reconstruction principles. The uses Infrared Structured Light Reconstruction to enable immersive interaction for users within environment. focus of the research is explore application technology in developing an efficient VR system. Starting from background applications, current challenges, and potential advantages technology, establish theoretical foundation our study....
3D measurement of underwater targets could recover morphology objects/scenes in water, which has extensive application prospects the fields submarine map drawing, resource exploration, and marine archaeology et al. reconstruction based on stereoscopic vision is playing a more important role field due to its incomparable advantages, such as high automation, rapid accuracy non-contact. However, target detection limited by complex environment, absorption scattering light water so on, will...
Measurement of specular surface shape is still a challenging task for existing 3D scanning techniques. In this paper, an accurate phase measuring deflectometry method investigated reconstruction mirror surfaces. To calibrate the system parameters, standard flat used. By reflected checkerboard pattern displayed by LCD screen, camera intrinsic parameters are calibrated firstly. The last orientation calibration aligned as reference plane. With such manner, lens distortion and position can be...