1. Intorduction

강의 소개 :

Hello, and welcome to this course, Multi-object Tracking for Automotive Systems. I'm Lennart Svensson, professor of Signal Processing at Chalmers University of Technology working with machine learning, Bayesian statistics, and multi-object tracking. I'm personally very excited about doing this course with you first because multi-object tracking is a topic which is quite challenging to learn on your own, and second, because the field has matured significantly over recent years. I hope that you will find our course an easy way to obtain a solid understanding of all the main components in the field. And I'm Karl Granstrom. And I'm a researcher and teacher at Chalmers University of Technology. And I've been working on different object tracking problems for about 10 years with a focus on extended object tracking. And during this time, I've been part of many interesting collaborations and projects where tracking has been applied to data from cameras, and from LIDARs and radars. And I think it's a truly fascinating topic. And I look forward to teaching this course. As you know, the automotive and transportation sectors are undergoing a transformation. Autonomous driving is destined to fundamentally change the way we think of cars and transportation. To be safe, autonomous vehicles must be capable of perceiving their surroundings and all objects that move around. In a nutshell, multi-object tracking is about the accurate perception of the driving environment. And this is a key enabling technology for any self-driving vehicle system. The course is divided into six lectures. And we will teach three lecturers each. In the first lecture, we will show some motivational examples of multi-object tracking applications. We will define the fundamentals of tracking, and we will discuss what makes tracking a challenging problem. Lecture two is about single-object tracking in clutter. This is where we learn about the basics of object tracking. In particular, we learn about standard models for measurements, when the measurements may contain both object measurements and false measurements. And we study a few simple algorithms for tracking. In the third lecture, we will learn about tracking a known number of multiple objects. And the number of objects is assumed known here to give a gradual increase in the complexity of the problems that we face. So the basic modeling of this problem will be explained, and we will present some different types of algorithms for tracking a known number of objects. In lecture four, we relax the assumption that the number of objects is known, and we introduce random finite sets as an elegant tool to represent uncertainties in both the number of objects that are present and the states of these objects. Random finite sets provide us with a unified framework to model and handle all aspects of multi-object tracking and have become central to the field of multi-object tracking. In the fifth lecture, we will finally arrive where we want to be, and that is tracking an unknown and time-varying number of objects. So we will use the random finite set models, and we'll present and discuss different algorithms for multiple-object tracking. And this includes recent advances like Poisson Multi-Bernoulli Mixture filters, and delta Generalized Labeled Multi-Bernoulli filters. Finally in the last lecture, we end by briefly introducing three topics that we believe are important for multi-object tracking in an automotive context, which are deep learning, sets of trajectories, and extended-object tracking. As an example, there are a few videos about deep learning, but rather than teaching you how to do deep learning, which is beyond the scope of this course, we highlight the importance of deep learning in multi-object tracking and discuss the role it may play in the future. Please note that this course is a sequel to Sensor Fusion and Non-linear Filtering for Automotive Systems. So therefore, unless you already have a solid background in sensor fusion and non-linear filtering, we strongly recommend that you first complete that course. Of course, it's available via edX. When questions or thoughts pop up about the content, exercises, or assignments, don't hesitate to make a post in the discussion forum. We have seen many times that active participation in discussions about the material improves the learning. Again, most welcome to this course. We look forward to getting to know you. And good luck with your studies.