Visual Object Tracking

Back to Verticals page

In this vertical we advance Visual Object Tracking (VOT) technology through cutting-edge machine vision for improved accuracy and real-time monitoring.

Different surveys of the VOT are performed:

Related Publications

● Handcrafted and Deep Trackers: Recent Visual Object Tracking Approaches and Trends, ACM Computing Surveys, 2019
● Tracking Noisy Targets: A Review of Recent Object Tracking Approaches, arXiv, 2018
● Comparative Study of ECO and CFNet Trackers in Noisy Environment, ICNGC, 2017

● Improving Object Tracking by Added Noise and Channel Attention, MDPI Sensors, 2020
● Deep Siamese Networks Towards Robust Visual Tracking, Book: Visual Object Tracking in the Deep Neural Networks, 2019

a. Information Fusion for Visual Object Tracking (VOT)

This project aims to develop novel information fusion strategies, incorporating a common low-rank subspace for the fusion of diverse features and tracker responses for VOT. This strategy focuses on interpreting response maps as smoothly varying functions represented by individual low-rank matrices, effectively eliminating noise and sparse artifacts. Additionally, common low-rank subspaces are estimated, ensuring proximity to each subspace and enhancing fusion efficiency.

Related Publications

● A Novel Algorithm Based on a Common Subspace Fusion for Visual Object Tracking, Access, 2022
● Robust Tracking via Feature Enrichment and Overlap Maximization, IW-FCV, 2021
● Learning soft mask-based feature fusion with channel and spatial attention for robust visual object tracking, MDPI Sensors, 2020

b Video Object Segmentation (VOS)

In this project, advancements in computer vision will be explored, with a specific focus on video object segmentation and moving object segmentation. New methods will be introduced to enhance VOS by incorporating context. Semi- supervised VOS will also be investigated through guided feature learning. The study aims to provide valuable insights and solutions to contemporary challenges in VOS.

Related Publications

● Video Object Segmentation Based on Guided Feature Transfer Learning, IW-FCV, 2022
● 4G-VOS: Video Object Segmentation using guided context embedding, Knowledge-Based Systems, 2021
● Video Object Segmentation using Guided Feature and Directional Deep Appearance Learning, CVPR Workshop, 2020

● Unsupervised RGBD Video Object Segmentation Using GANs, ACCV Workshop, 2018

c. Graph-Regularized Discriminative Correlation Filter for VOT

In this project, we will address the challenges of visual object tracking by introducing hierarchical spatiotemporal graph-regularized correlation filters. The target sample is decomposed into deep channels, and spatial and temporal graphs are constructed to capture the spatial and temporal structures of the target object. These aim is to improve robustness and performance in challenging tracking scenarios.

Related Publications

● Hierarchical Spatiotemporal Graph Regularized Discriminative Correlation Filter for Visual Object Tracking, IEEE Transactions on Cybernetics, 2021
● Spatiotemporal Low-rank Modeling for Complex Scene Background Initialization, IEEE Transactions on Circuits and Systems for Video Technology, 2018

d. Robust Structural Low-Rank Tracking

In this project we explore deep neural networks for visual object tracking that will take advantage of two stacked patches to regress similarity and dissimilarity scores in a single evaluation simultaneously. These networks will process concatenated depth-wise image patches to exploit structural differences between the inputs for improved accuracy.

Related Publications

● Structural Low-Rank Tracking, AVSS, 2019

● Improving Object Tracking by Added Noise and Channel Attention, MDPI Sensors, 2020

● Robust Structural Low-Rank Tracking, IEEE Transactions on Image Processing (TIP), 2020

● Convolutional Neural Network with Structural Input for Visual Object Tracking, ACM/SIGAPP Symposium on Applied Computing, 2019

e. Complete Moving Object Detection:

This project will focus on the crucial task of complete moving object detection in computer vision applications, addressing challenges such as dynamic backgrounds, camouflage, illumination variations, and noise. This study introduces novel approaches using a conditional Generative Adversarial Network (cGAN), specifically conditioned on non-occluded moving object pixels during training.

Related Publications

● Complete Moving Object Detection in the Context of Robust Subspace Learning, ICCV Workshop, 2019

f. Improving Underwater Visual Object Tracking

In this project, we focus on advancing Underwater Visual Object Tracking (UVOT) through the creation of new datasets and introducing novel tracker enhancement methods. We propose innovative underwater image enhancement algorithms explicitly tailored to elevate the quality of tracking in underwater environments, paving the way for improved UVOT capabilities.

Related Publications

● Improving Underwater Visual Tracking With a Large Scale Dataset and Image Enhancement, arXiv, 2023

Visual Object Tracking

AI for Anomalous Events Detection

Self-supervised and Unsupervised Learning

Back to Verticals page