Predictive 3D Perception for Intelligent Vehicles (half-day)
Aim of the Tutorials and Intended Audience
3D perception, the ability to perceive depth and spatial relationships in the world, is fundamental to human cognition and holds immense potential across various sensing domains, such as intelligent vehicles. The emergence of deep learning-based techniques offers a compelling alternative, enabling 3D vision from monocular camera inputs without additional hardware modifications.
This tutorial will explore the principles and applications of traditional 3D sensing and computer vision methods for intelligent vehicles. Subsequently, we will introduce predictive 3D sensing based on deep learning, covering fundamental concepts, common architectures, and training data requirements. We will use simultaneous localization and mapping (SLAM) and autonomous driving as illustrations. This tutorial will also discuss issues such as compatibility with monocular cameras and seamless integration into existing sensor systems without requiring additional hardware modifications.
Expected Participation
This tutorial is designed for a broad spectrum of professionals, including researchers, engineers, and practitioners, who are intrigued by sensing, computer vision, machine learning, and intelligent vehicles. This tutorial welcomes individuals eager to deepen their understanding of novel approaches for soft sensing based on generative AI and explore the transformative potential of generative/predictive AI in the sensors, robotics, and intelligent transportation systems domains.
Experienced professionals seeking to expand their knowledge and stay updated on the latest advancements in generative AI and intelligent vehicles will find the tutorial beneficial. Newcomers keen on delving into the intersection of computer vision, machine learning, and intelligent vehicles will discover valuable insights and practical guidance. The tutorial aims to foster a collaborative learning environment
conducive to knowledge exchange and skill enhancement by accommodating participants with diverse backgrounds and expertise levels.
An Outline of the Tutorial Focus
1. Introduction to SLAM and Intelligent Vehicles
Definition and importance in various applications.
Overview of current monocular and RGB-D SLAM techniques and their limitations.
2. State-of-the-Art 3D Perception Methods:
RGB-D sensors: Principles and applications.
Stereo sensors: How they work and their applications.
Time-of-flight sensors: Understanding the technology and its applications.
LIDAR: Overview of principles and applications.
3. Predictive 3D Sensing:
Basics of deep learning for depth estimation.
Deep-learning architectures are commonly used for depth prediction tasks.
Training data requirements and challenges.
4. RGB Predictive Depth SLAM
Compatibility with monocular cameras: Leveraging a single camera for depth estimation.
Integration with existing sensor systems: No additional hardware modifications are required.
System development and comparison with monocular SLAM and RGB-D SLAM
5. Applications
Intelligent vehicles
Unmanned Aerial Vehicles and Obstacle Avoidance
6. Hands-On Session:
Demonstration of depth sensing using deep learning models.
Practical exercises on implementing depth estimation algorithms.
Q&A and discussion on implementation challenges and best practices.
The Format and Proposed Schedule of the Session
The proposed tutorial will have two parts. Each part is about 60-70 minutes, with a healthy break in between. The first part will cover topics 1-3 in the outline, and the second will cover the rest. The format will be a lecture with discussion and demonstration.
Information about the Presenters/Organizers.
Henry Leung is the Schulich Industrial Chair Professor of the Department of Electrical and Software Engineering of the University of Calgary. Before joining U of C, he was a defense scientist with the Department of National Defence (DND) of Canada. His research interests include information fusion, machine learning, IoT, data analytics, robotics, signal, and image processing. He has published more than 350 journal papers and 250 conference papers. He is an associate editor of journals such as Scientific Reports, IEEE Emerging Topics on Circuits and Systems, and Journal of Sensors. He is the Springer book series editor on “Information Fusion and Data Science.” He has given tutorials at various major conferences such as the IEEE Sensors Conference (2022), IEEE System, Man and Cybernetics (SMC) Conference (2023), and ISPRS Geospatial Week (2023) on AI, machine learning, and its application to sensors, autonomous systems, and remote sensing. He has delivered keynote speeches on 3D perception and autonomous vehicles at four recent international conferences related to the proposed
tutorial. He is a Fellow of IEEE, SPIE, Engineering Institute of Canada (EIC), and the Canadian Academy of Engineering (CAE).
Gayan Brahmanage holds a BSc in Engineering from the University of Moratuwa, Sri Lanka, and both MSc and PhD degrees in Electrical and Software Engineering from the University of Calgary, Canada. His expertise includes SLAM, Robotics, and 3D perception. He is actively involved in multiple development projects focusing on robotics and 3D perception.
Names of the Organizers and Contact Person
Dr. Henry Leung
Schulich Industry Research Chair Professor
Department of Electrical and Software Engineering
University of Calgary
2500 University Drive NW
Calgary, Alberta, Canada T2N 1N4
Email: leungh@ucalgary.ca
Dr. Gayan Brahmanage
Department of Electrical and Software Engineering
University of Calgary
2500 University Drive NW
Calgary, Alberta, Canada T2N 1N4
Email: gayan.brahmanage@ucalgary.ca