A5066468110- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Human Pose and Action Recognition
- Image Retrieval and Classification Techniques
- Robotics and Sensor-Based Localization
- Advanced Vision and Imaging
- Anomaly Detection Techniques and Applications
- Gait Recognition and Analysis
- Face recognition and analysis
- Radiomics and Machine Learning in Medical Imaging
- Remote-Sensing Image Classification
- 3D Shape Modeling and Analysis
- Image Enhancement Techniques
- Visual Attention and Saliency Detection
- Face and Expression Recognition
- Biometric Identification and Security
- AI in cancer detection
- 3D Surveying and Cultural Heritage
- Hand Gesture Recognition Systems
- Multimodal Machine Learning Applications
- Automated Road and Building Extraction
- Remote Sensing and Land Use
- Image and Object Detection Techniques
- Video Analysis and Summarization
Xiamen University
2015-2024
Stanford University
2024
Chinese PLA General Hospital
2024
Asahi Kasei (Japan)
2020
Xiamen University of Technology
2011-2019
Hebei Normal University
2013
Intelligent Health (United Kingdom)
2012
Brainlike (United States)
2012
3D point cloud-based place recognition is highly demanded by autonomous driving in GPS-challenged environments and serves as an essential component (i.e. loop-closure detection) lidar-based SLAM systems. This paper proposes a novel approach, named NDT-Transformer, for real-time large-scale using clouds. Specifically, Normal Distribution Transform (NDT) representation employed to condense the raw, dense cloud probabilistic distributions (NDT cells) provide geometrical shape description. Then...
In this paper, we investigate the possibility of monitoring traffic without using any motion features. The goal our system is to process videos with ultra-low frame rate, i.e. for which reliable features cannot be computed. work, how 2D spatial combined a machine learning method can assess conditions such as fluid traffic, dense and jam. underlying hypothesis that ought validate images are heavily characterized by their textures. perspective, tested different texture methods see accurate an...
In this paper, we investigate the possibility of monitoring highway traffic based on videos whose frame rate is too low to accurately estimate motion features. The goal proposed method recognize conditions instead measuring them, as usually case. main advantage our approach comes from its ability process low-frame-rate for which features cannot be estimated. Our takes highly redundant nature scenes that are pictured a top-down perspective showing vehicles predominant asphalted road...
This paper proposes a self-supervised learned local detector and descriptor, called EventPoint, for event stream/camera tracking registration. Event-based cameras have grown in popularity because of their biological inspiration low power consumption. Despite this, applying features directly to the stream is difficult due its peculiar data structure. We propose new time-surface-like representation method Ten-code. The processed by Tencode can obtain pixel-level positioning interest points...
Pedestrian detection is an active area of research with challenge in computer vision.This study conducts a detailed survey on state-of-the-art pedestrian methods from 2005 to 2011,focusing the two most important problems:feature extraction,the classification and localization.We divided these into different categories;pedestrian features are three subcategories:low-level feature,learning-based feature hybrid feature.On other hand,classification localization also sub-categories:sliding window...
A feature, named structured local binary Haar pattern (SLBHP), is proposed for pixel-based graphics retrieval. The SLBHP a hybrid of (LBP) and wavelet. encodes the polarity rather than magnitude difference between accumulated grey values adjacent rectangles. relationships are then considered as value in LBP. Experimental results on retrieval show that discriminative power superior to those using edge points, LBP even noisy conditions.
We present an efficient scene layout aware object detection method for traffic surveillance. Given input image, our approach first estimates its by transferring annotations in a large dataset to the target image based on nonparametric label transfer. The transferred are then integrated with hypotheses generated state-of-the-art detectors. propose approximate nearest neighbor search scheme inference estimation. Experiments verified that this simple and provides consistent performance...
Summary Imbalanced samples are widespread, which impairs the generalization and fairness of models. Semi‐supervised learning can overcome deficiency rare labeled samples, but it is challenging to select high‐quality pseudo‐label data. Unlike discrete labels that be matched one‐to‐one with points on a numerical axis, in regression tasks consecutive cannot directly chosen. Besides, distribution unlabeled data imbalanced, easily leads an imbalanced data, exacerbating imbalance semi‐supervised...
Accurate segmentation of breast tumors in dynamic contrast-enhanced magnetic resonance images (DCE-MRI) is critical for early diagnosis cancer. However, this task remains challenging due to the wide range tumor sizes, shapes, and appearances. Additionally, complexity further compounded by high dimensionality ill-posed artifacts present DCE-MRI data. Furthermore, accurately modeling features sequences presents a challenge that hinders effective representation essential characteristics....