Dr. Zheng (Thomas) Tang

Dr. Zheng (Thomas) Tang

Senior Deep Learning Engineer at NVIDIA, expert in computer vision and machine learning

Robust Video Object Tracking via Camera Self-Calibration - My Ph.D. Defense Experience

1 minute read

Published:

I am pleased to announce that I have successfully passed my Ph.D. Dissertation Defense in the Department of Electrical and Computer Engineering (ECE) at the University of Washington (UW). This has been a long and challenging journey, and I am grateful for the support of my supervisory committee, colleagues, sponsors, family, friends, and my faith.

My dissertation, entitled “Robust Video Object Tracking via Camera Self-Calibration,” focuses on improving the accuracy and robustness of video object tracking. Video object tracking is a crucial task in computer vision that involves identifying and tracking an object of interest in a video sequence. It has numerous applications, such as surveillance, autonomous driving, and virtual reality.

However, video object tracking is a challenging task due to various factors such as occlusions, illumination changes, and camera motion. In my dissertation, I propose a novel approach to video object tracking that utilizes camera self-calibration to enhance its robustness and accuracy. Camera self-calibration is the process of estimating the intrinsic and extrinsic parameters of a camera from a set of images without the need for external calibration tools.

My supervisory committee consisted of Prof. Jenq-Neng Hwang (Chair), Prof. Kenneth P. Bube (GSR), Prof. Linda G. Shapiro, Prof. Ming-Ting Sun, and Prof. Fa-Long Luo. I am grateful for their guidance, feedback, and support throughout my Ph.D. journey. Their expertise and insights were invaluable in shaping my research and helping me navigate the challenges of graduate school.

In conclusion, I am thrilled to have successfully defended my Ph.D. dissertation and am excited to continue exploring the exciting field of computer vision. Once again, I would like to express my heartfelt gratitude to my committee, colleagues, sponsors, family, friends, and my faith for their unwavering support and encouragement throughout my Ph.D. journey.

 Photo

You May Also Enjoy

Simplifying Industrial Automation with Real-Time Multi-Camera AI

2 minute read

Published:

In the heart of the industrial automation revolution, the Metropolis multi-camera tracking system emerges as a beacon of innovation, seamlessly integrating with NVIDIA’s AI suite to redefine efficiency and safety in complex industrial settings. Developed by a pioneering software engineer, Metropolis creates a real-time, comprehensive map from hundreds of camera feeds, guiding autonomous mobile robots through intricate environments with unparalleled precision. This fusion of real-time AI and digital twin technology not only showcases the potential to drastically reduce operational downtime but also marks a significant leap forward in the quest for smarter, more responsive industrial ecosystems. Through this lens, we glimpse the future of automation, where digital precision and human ingenuity converge to create harmonious, highly optimized workplaces.

Introducing Metropolis AI Workflows & Microservices 1.0: Revolutionizing Multi-Camera Tracking

3 minute read

Published:

The Metropolis AI Workflows & Microservices 1.0 is officially live and is set to revolutionize the way enterprises and our ecosystem approach centralized perception across an array of matrixed sensors. One of the most exciting features of this release is the Multi-Camera Tracking app, which I had the pleasure of developing. The app is a reference architecture for video analytics applications that tracks people across multiple cameras and provides the counts of unique people seen over time. This is also known as Multi-Target Multi-Camera (MTMC) tracking.

My Work on the NVIDIA TAO Toolkit 4.0: People Re-Identification and Pose-Based Action Recognition Networks

2 minute read

Published:

NVIDIA has recently announced the release of the TAO Toolkit 4.0, which includes several exciting new features and enhancements. As a developer who has contributed to the toolkit, I’m thrilled to share my experience working on the people re-identification and pose-based action recognition networks, as well as the end-to-end video analytics pipelines on the Triton Inference Server.