Intensive-care unit patients monitoring by computer vision system

Abstract

In this project, we propose an automatic computer vision system for patient monitoring at the Intensive-Care Unit (ICU). These patients require constant monitoring and, due to the high costs associated to equipment and staff necessary, the design of an automatic system would be helpful. Depth imaging technology has advanced dramatically over the last few years, finally reaching a consumer price point with the launch of Kinect. These depth images are not affected by the lighting conditions and provide us a good vision, even without any light, so we can monitorize the patients 24 hours a day. In this project, we worked on two of the parts of the object detection systems: the descriptor and classifier. Concerning the descriptor, we analyzed the performance of one of the most used descriptors for object detection in RGB images, the Histogram of Oriented Gradients, and we have proposed a descriptor designed for depth images. It is shown that the combination of these two descriptors increases system accuracy. As to the detection, we have done various tests. We analyzed the detection of patient body parts separately, and we have used a model where the patient is divided into multiple parts and each part is modeled with a set of templates, demonstrating that the use of a model helps to improve detection.