A5084204275- Gaze Tracking and Assistive Technology
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Industrial Vision Systems and Defect Detection
- Robotics and Automated Systems
- Domain Adaptation and Few-Shot Learning
- Video Surveillance and Tracking Methods
- Image and Object Detection Techniques
- 3D Surveying and Cultural Heritage
- Advanced Computing and Algorithms
- Image Processing Techniques and Applications
- Robotics and Sensor-Based Localization
- Handwritten Text Recognition Techniques
- Ergonomics and Musculoskeletal Disorders
- Human Pose and Action Recognition
- Hand Gesture Recognition Systems
- Vehicle License Plate Recognition
- Image Enhancement Techniques
- Glaucoma and retinal disorders
- Sleep and Work-Related Fatigue
- Anomaly Detection Techniques and Applications
- Color perception and design
- Advanced Measurement and Detection Methods
- Dental Radiography and Imaging
- Face recognition and analysis
Shanghai University
2014-2024
Beijing Institute of Radio Metrology and Measurement
2021
Peking University
2018-2020
State Key Laboratory of Industrial Control Technology
2011
University of Science and Technology Liaoning
2003
We present MMDetection, an object detection toolbox that contains a rich set of and instance segmentation methods as well related components modules. The started from codebase MMDet team who won the track COCO Challenge 2018. It gradually evolves into unified platform covers many popular contemporary not only includes training inference codes, but also provides weights for more than 200 network models. believe this is by far most complete toolbox. In paper, we introduce various features...
Feature pyramids are widely exploited by both the state-of-the-art one-stage object detectors (e.g., DSSD, RetinaNet, RefineDet) and two-stage Mask RCNN, DetNet) to alleviate problem arising from scale variation across instances. Although these with feature achieve encouraging results, they have some limitations due that only simply construct pyramid according inherent multiscale, pyramidal architecture of backbones which originally designed for classification task. Newly, in this work, we...
In existing CNN based detectors, the backbone network is a very important component for basic feature1 extraction, and performance of detectors highly depends on it. this paper, we aim to achieve better detection by building more powerful from ones like ResNet ResNeXt. Specifically, propose novel strategy assembling multiple identical backbones composite connections between adjacent backbones, form named Composite Backbone Network (CBNet). way, CBNet iteratively feeds output features...
The convolution operation suffers from a limited receptive filed, while global modeling is fundamental to dense prediction tasks, such as semantic segmentation. In this paper, we apply graph into the segmentation task and propose an improved Laplacian. reasoning directly performed in original feature space organized spatial pyramid. Different existing methods, our Laplacian data-dependent introduce attention diagonal matrix learn better distance metric. It gets rid of projecting...
The panoptic segmentation task requires a unified result from semantic and instance outputs that may contain overlaps. However, current studies widely ignore modeling In this study, we aim to model overlap relations among instances resolve them for segmentation. Inspired by scene graph representation, formulate the overlapping problem as simplified case, named graph. We leverage each object's category, geometry appearance features perform relational embedding, output relation matrix encodes...
The captured underwater images suffer from color cast and haze effect caused by absorption scattering. These interdependent phenomena jointly degrade images, resulting in failure of autonomous machines to recognize image contents. Most existing learning-based methods for enhancement (UIE) treat the degraded process as a whole ignore interaction between correction dehazing. Thus, they often obtain unnatural results. To this end, we propose novel joint network optimize results dehazing...
Since underwater images are seriously degraded due to the attenuation of light, artificial light (AL) is often used assist photography in underwater. However, normal imaging process changed by AL. It observed that AL source typically alters condition a large extent, resulting non-uniform illumination images. In addition, color distortion area affected little because close object suffers attenuation. most existing image enhancement algorithms ignore this phenomenon. their results, areas tend...
The ability to detect small objects and the speed of object detector are very important for application autonomous driving, in this paper, we propose an effective yet efficient one-stage detector, which gained second place Road Object Detection competition CVPR2018 workshop - Workshop Autonomous Driving(WAD). proposed inherits architecture SSD introduces a novel Comprehensive Feature Enhancement(CFE) module into it. Experimental results on dataset as well MSCOCO demonstrate that (named...
Interior wall painting is widely used in construction, installation repair, decoration, and other construction industry. By replacing human manual operation, robotic developed to improve efficiency reduce the cost. This paper proposes a parameters estimation scheme for process. The parameter system built on single camera, laser level, four distance sensors. camera's intrinsic are obtained by calibration method. relative position between robot through single-image measurement. plane equation...
Abstract Slow detection of redundant objects and low accuracy in assembly lines, particularly the setting civil aircraft assembly, are tough challenging problems. To address these issues, a object method based on computer vision augmented reality (AR) smart glasses is proposed this paper. The uses AR as image collection hardware takes live collected by camera input deep learning machine model. model, Feature Pyramid Networks-CenterNet, inspired CenterNet combined with multi-scale feature...
Many outdoor activities, such as bicycling, driving, and riding segways can be modeled rigid bodies on a moving platform. In this study, an attitude estimation scheme is established for the body-moving platform system by using two gyroscopes relative measurements between body The proposed estimates drift-free attitudes of partial absolute without any global information or reference. kinematic model plays important role to obtain angles. To illustrate this, compared another design with...
In this report, our approach to tackling the task of ActivityNet 2018 Kinetics-600 challenge is described in detail. Though spatial-temporal modelling methods, which adopt either such end-to-end framework as I3D \cite{i3d} or two-stage frameworks (i.e., CNN+RNN), have been proposed existing state-of-the-arts for task, video far from being well solved. challenge, we propose network (StNet) better joint and comprehensively understanding. Besides, given that multi-modal information contained...
Wood defect detection has an important influence on the automation of wood industry. In view complexity segmentation, this paper proposes hybrid algorithm genetic and particle swarm optimization algorithm. Firstly, contrast image is enhanced by linear transformation function. Then, applying swarm-genetic respectively for finally morphological processing performed. The result shows that a better more stable effect detecting defects compared with
In existing CNN based detectors, the backbone network is a very important component for basic feature extraction, and performance of detectors highly depends on it. this paper, we aim to achieve better detection by building more powerful from backbones like ResNet ResNeXt. Specifically, propose novel strategy assembling multiple identical composite connections between adjacent backbones, form named Composite Backbone Network (CBNet). way, CBNet iteratively feeds output features previous...
Slip is one of the major causes for falls during human walking. Providing external knee assistive torque can potentially help people perform successful slip recovery and avoid serious injuries. In this paper, we present a design characterization wearable robotic device slip-induced fall prevention. The consists set compliant components with impedance feedback control. bench performance evaluations show that capable tracking angle profile walking gait therefore be worn regular without...
Natural human-robot interaction plays an important role in effective nursing services system provided by service robots for the elderly and disabled people. This paper proposed a multimodal "human-robot integration" collaboration system, set up shared interface between human robot. Consequently, Users can naturally communicate retrieve information from collaborative with multimodality(e.g. head gesture, eye gaze) interactive dialogue approach. By this way, making fully understand human's...
Scene text recognition is very challenging due to the complex background, low resolution, perspective distortion and curved placement, etc. Most of state-of-the-art methods adopt attention-based encoder-decoder framework, usually get suboptimal performance for images misalignment between attention region target character region. In this paper, a novel module, named Gated Cascade Attention Module (GCAM), proposed increase alignment precision in cascade way. Moreover, channel spatial module...
Feature pyramids are widely exploited by both the state-of-the-art one-stage object detectors (e.g., DSSD, RetinaNet, RefineDet) and two-stage Mask R-CNN, DetNet) to alleviate problem arising from scale variation across instances. Although these with feature achieve encouraging results, they have some limitations due that only simply construct pyramid according inherent multi-scale, pyramidal architecture of backbones which actually designed for classification task. Newly, in this work, we...