Human motion tracking-by-detection using point cloud data from multiple depth sensors

点击翻译按钮获取中文摘要

In research and practice, using computers to track people is important for many reasons including security, healthcare delivery, computer interfaces, and video gaming. Colour video cameras, and more recently colour cameras that can detect how far objects are from the cameras, have been used to track people in indoor environments without the use of any special equipment (like wearable sensors or sensors placed around the room). However, these only work well in ideal indoor environments rooms with objects that don't move very often, and rooms without a lot of clutter. In cases where rooms are dynamic (e.g., in hospital settings) or rooms are cluttered (e.g., busy homes), these methods of tracking do not work well. In these more challenging indoor environments, these systems often do not work because the single camera cannot see enough of the room to track the people or because the people are blocked by the clutter in the room. ***This work wants to find a way to reliably track people in more challenging indoor environments. Adding more cameras is a possible solution, but programming a computer to track people with more than one camera is difficult. Currently, images from more than one camera can only reliably be put together if the images are almost the same (e.g., looking at the room from almost the same angle). Even if the cameras are looking at the same person or part of the room from different angles, it is very hard to put together the images. The proposed work will partially align images from more than one camera using the natural motion of people as they walk around a room. Once partially aligned, existing ways of reliably putting together images can then be used. After the images from the cameras are put together into one complete image, the actions of the people will be tracked to identify what they are doing. However, so far tracking what people are doing using an image made from more than one camera has been hard. This work will also develop a new way of tracking what people are doing from this combined image in challenging indoor environments.***Tracking people in more challenging indoor environments will allow helpful technologies to be used in settings like hospitals, long-term care homes, and even in the homes of healthy people. In this way, technology can help us in many ways like improving the way we get help in hospitals, making us safer, making it easier to use computers, and making video games more fun and realistic to play. This type of research is also very timely and important to other researchers, at the state-of-the-art of the field, because it will allow others to easily combine images from multiple cameras and track what people are doing in most indoor environments. This research will result in the creation of a prototype system that will automatically start tracking people in an indoor environment after users place cameras anywhere they want around a room.**