A5048871510- Video Surveillance and Tracking Methods
- Fire Detection and Safety Systems
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Human Pose and Action Recognition
- Anomaly Detection Techniques and Applications
- Visual Attention and Saliency Detection
- Air Quality Monitoring and Forecasting
- Infrared Target Detection Methodologies
- COVID-19 diagnosis using AI
- Advanced Image and Video Retrieval Techniques
- Artificial Intelligence in Healthcare and Education
- IoT-based Smart Home Systems
- Advanced Chemical Sensor Technologies
- Radiomics and Machine Learning in Medical Imaging
- COVID-19 epidemiological studies
- Power Systems and Technologies
- Image Enhancement Techniques
- Digital Imaging for Blood Diseases
University of British Columbia
2020-2022
Okanagan University College
2022
Kelowna General Hospital
2021
Memorial University of Newfoundland
2018-2020
University of Waterloo
2020
Indian Institute of Technology Roorkee
2020
Lakehead University
2020
Queen's University
2020
National Research Council Canada
2020
The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by VOT initiative. Results of 71 trackers are presented; many state-of-the-art published at major computer vision conferences or in journals recent years. was composed four sub-challenges focusing on different tracking domains: (i) VOT-ST2021 focused short-term RGB, (ii) VOT-RT2021 "real-time" (iii) VOT-LT2021 long-term tracking, namely coping with target disappearance and reappearance...
Balancing the trade-off between real-time performance and accuracy in object tracking is a major challenge. In this paper, novel dynamic policy gradient Agent-Environment architecture with Siamese network (DP-Siam) proposed to train tracker increase expected average overlap while performing real-time. DP-Siam trained offline reinforcement learning produce continuous action that predicts optimal location. has consists of three networks: an Agent predict state (bounding box) being tracked,...
Abstract Classifying and analyzing human cells is a lengthy procedure, often involving trained professional. In an attempt to expedite this process, active area of research involves automating cell classification through use deep learning-based techniques. practice, large amount data required accurately train these learning models. However, due the sparse datasets currently available, performance models typically low. This study investigates feasibility using few-shot techniques mitigate...
The COVID-19 pandemic has been deemed a global health pandemic. early detection of is key to combating its outbreak and could help bring this an end. One the biggest challenges in accurate testing for disease. Utilizing power Convolutional Neural Networks (CNNs) detect from chest X-ray images can radiologists compare validate their results with automated system. In paper, we propose carefully designed network, dubbed CORONA-Net, that accurately images. CORONA-Net divided into two phases: (1)...
Background/Foreground separation is a fundamental and challenging task of many computer vision applications. The F-measure performance state-of-the-art models limited due to the lack fine details in predicted output (i.e., foreground object), labeled data. In this paper, we propose background/foreground model based on transformer that has higher learning capacity than convolutional neural networks. trained using self-supervised leverage data learn strong object representation invariant...
Visual tracking is a difficult and challenging problem, for numerous reasons such as small object size, pose angle variations, occlusion, camera motion. Object has many real-world applications surveillance systems, moving organs in medical imaging, robotics. Traditional methods lack recovery mechanism that can be used situations when the tracked objects drift away from ground truth. In this paper, we propose novel framework based on composite reporter mechanism. The tracks using different...
Recently, the COVID-19 pandemic has affected world and spread in majority of countries. To decrease number infections, experts suggested people practice social distancing by maintaining a distance six feet apart. It is hard to monitor this restriction only traditional surveillance system. Existing methods used deep learning tackle problem designing Deep Convolutional Neural Network (DCNN). However, these do not accommodate for low-power systems such as Internet-of-Things-based devices. In...
We propose NullSpaceNet, a novel network that maps from the pixel-level image to joint-nullspace, as opposed traditional feature space. The features in proposed learned joint-nullspace have clearer interpretation and are more separable. NullSpaceNet ensures all input images belong same class collapsed into one point this new of different classes points with high separation margins. Moreover, differentiable loss function is has closed-form solution no free parameters. architecture consists...
Convolutional Siamese neural networks have been recently used to track objects using deep features. architecture can achieve real time speed, however it is still difficult find a that maintains the generalization capability, high accuracy and speed while decreasing number of shared parameters especially when very deep. Furthermore, conventional usually processes one local neighborhood at time, which makes appearance model non-robust changes. To overcome these two problems, this paper...
Vehicle object detection is a fundamental task in computer vision. Most modern classifiers and trackers are built upon the detectors. For example, self-driving cars use on low-power devices to capture information from surrounding environment. Currently, uses huge amount of labelled data train detector. Moreover, these detectors designed for high-end hardware (i.e., GPUs) cannot be used devices. In this paper, we propose DETECTren, novel detector that self-supervised learning leverage both...
We propose NullSpaceNet, a novel network that maps from the pixel level input to joint-nullspace (as opposed traditional feature space), where newly learned features have clearer interpretation and are more separable. NullSpaceNet ensures all inputs same class collapsed into one point in this new joint-nullspace, different classes points with high separation margins. Moreover, differentiable loss function is proposed has closed-form solution no free-parameters. exhibits superior performance...
COVID-19 affects everyone on a daily-basis causing adjustments in which society functions. One of these major is the need to measure how well people distance from each other, that referred as social distancing. Previous work automate distancing violations does not take into consideration exceptions minimum guidelines. In this paper, we propose GroupNet, novel multi-object tracking violation detector through addition group detection reduce number false positives are currently missed existing...
Visual object tracking is a fundamental task in the field of computer vision. Recently, Siamese trackers have achieved state-of-the-art performance on recent benchmarks. However, do not fully utilize semantic and objectness information from pre-trained networks that been trained image classification task. Furthermore, architecture sparsely activated by category label which leads to unnecessary calculations overfitting. In this paper, we propose learn Domain-Aware, utilizing while producing...